Content uploaded by Saurabh Vishnubhakat
Author content
All content in this area was uploaded by Saurabh Vishnubhakat on Jul 13, 2017
Content may be subject to copyright.
Harvard Journal of Law & Technology
Volume 29, Number 1 Fall 2015
WHEN BIOPHARMA MEETS SOFTWARE:
BIOINFORMATICS AT THE PATENT OFFICE
Saurabh Vishnubhakat & Arti K. Rai*
TABLE OF CONTENTS
I. INTRODUCTION .............................................................................. 206!
II. QUALITY IN THE PATENT OFFICE ................................................. 208!
A. Improper Application of Patent Law Standards ....................... 211!
B. Lax Patentability Standards ..................................................... 212!
C. Quality and Technology-Specificity ......................................... 213!
III. MEASURING PATENT QUALITY ................................................... 217!
A. Citations and Knowledge Transfer ........................................... 218!
B. Private Value ............................................................................ 219!
C. Legal Validity ........................................................................... 220!
D. Examiner Characteristics ......................................................... 221!
IV. BIOINFORMATICS AT THE USPTO: AN EMPIRICAL VIEW ........... 223!
A. Technology and Comparison Group ........................................ 223!
B. Hypotheses About Examination ................................................ 226!
C. Results: Applications ................................................................ 228!
D. Results: Examination ............................................................... 230!
E. Discussion ................................................................................. 232!
1. Subject Matter ........................................................................ 232!
2. Written Description: Impacts on Notice ................................ 232!
3. Other Indicia of Notice .......................................................... 233!
4. Enablement and Nonobviousness .......................................... 236!
V. POLICY IMPLICATIONS FOR PATENT QUALITY AND
TEAM-BASED INNOVATION ........................................................... 237!
A. Patent Quality ........................................................................... 237!
* Saurabh Vishnubhakat is an Associate Professor of Law at the Texas A&M University
School of Law and a Fellow at the Duke Law Center for Innovation Policy; until June, 2015,
he was an Expert Advisor at the United States Patent and Trademark Office. Arti K. Rai is
the Elvin R. Latty Professor of Law at the Duke University School of Law. This research
was supported by the National Institutes of Health’s National Human Genome Research
Institute under Grant P50 HG003391. We are grateful to Bhaven Sampat and to Ronald
Mann for invaluable data and methodological discussion, to Alex Trzeciak and Tom Watson
for excellent research assistance, and to participants at the 2015 Works in Progress Intellec-
tual Property Colloquium for helpful comments. Jorge Contreras, Ben Jones, Rachel Sachs,
and Melissa Wasserman also provided helpful comments. The rejections for all of the patent
applications we looked at, rather than just the matched sample of 132, are available on Har-
vard Dataverse. The arguments in this writing are the authors’ and should not be imputed to
the USPTO or to any other organization.
206 Harvard Journal of Law & Technology [Vol. 29
B. Team-Based Innovation ............................................................ 237!
VI. CONCLUSION ............................................................................... 240
I. INTRODUCTION
In 1999, nine years after the National Institutes of Health’s
(“NIH”) National Center for Human Genome Research had published
its first joint research plan, the project to sequence the human genome
was operating at full force.1 The entire genome of a free-living organ-
ism, Haemophilus influenzae, had already been sequenced,2 and the
first full human chromosome sequence would be published that same
year.3 The joint announcement by President Bill Clinton and United
Kingdom Prime Minister Tony Blair of the so-called rough draft hu-
man genome would be made the following year.4
This burgeoning body of genomic knowledge required analytical
tools for parsing and manipulating it productively. Though such tools
had long existed in computer science and had even been applied to
research problems in the life sciences under the designation of bioin-
formatics, they had not yet been systematized into a formal disci-
pline.5 The specialization of such tools to manage the peculiar scope
and scale of genomic information marked the origin of bioinformatics
as a distinct discipline.6
Perhaps not surprisingly, the United States Patent and Trademark
Office (“USPTO”) had begun receiving a growing number of patent
applications for inventions in the field of bioinformatics.7 Based on
industry input, it projected many more in the coming years.8 In re-
1. A Brief History of the Human Genome Project, NIH NATIONAL HUMAN GENOME
RESEARCH INSTITUTE (Nov. 8, 2012), http://www.genome.gov/12011239
[http://perma.cc/B6GW-5X7G].
2. Robert D. Fleischmann et al., Whole-Genome Random Sequencing and Assembly of
Haemophilus Influenzae Rd, 269 SCIENCE 496, 496 (1995).
3. Ian Dunham et al., The DNA Sequence of Human Chromosome 22, 402 NATURE 489,
489 (1999).
4. Press Release, The White House Office of the Press Secretary, President Clinton An-
nounces the Completion of the First Survey of the Entire Human Genome: Hails Public and
Private Efforts Leading to This Historic Achievement (June 26, 2000),
http://archive.hhs.gov/news/press/2000pres/20000626.html [http://perma.cc/T2XQ-L8RH].
5. In 2000, NIH issued a working definition of bioinformatics which accounted for the
utility of the field in the expansive use of available biological information. See NIH Working
Definition of Bioinformatics and Computational Biology, NIH BIOMEDICAL INFORMATION
SCIENCE AND TECHNOLOGY INITIATIVE (July 17, 2000),
http://www.bisti.nih.gov/docs/CompuBioDef.pdf [http://perma.cc/RKW9-5G2J].
6. Jorge L. Contreras, Implementing Procedural Safeguards for the Development of Bio-
informatics Interoperability Standards, 39 N. KY. L. REV. 87, 87 (2012).
7. Douglas Steinberg, New PTO Unit Examines Bioinformatics Applications, THE
SCIENTIST (Nov. 27, 2000), http://www.the-scientist.com/?articles.view/articleNo/13144/
[https://perma.cc/C66V-FJHP].
8. Id.
No. 1] Bioinformatics at the Patent Office 207
sponse, by December of 1999, the USPTO had established a new art
unit to examine all bioinformatics applications in a consistent way.9
The art unit resides in USPTO Technology Center 1600,10 which ex-
amines inventions in biotechnology and organic chemistry11 and is
designated art unit (“AU”) 1631.12
From its earliest days, the patent examiners in AU 1631 had di-
verse expertise not only in the biological sciences, but also in physics
and electrical engineering and, most importantly, computer science.13
The USPTO considered the software and data processing patent cases
of the late 1990s14 directly relevant to patents on computing tools for
analyzing biological systems, and accordingly advised bioinformatics
inventors to draw lessons from the software invention guidelines in
the Manual of Patent Examining Procedure (“MPEP”).15 Thus the
rapid expansion of software patentability seen in that era directly af-
fected AU 1631.
However, the art unit was also located in Technology Center 1600,
which had just announced important, new, relatively strict examina-
tion guidelines on the so-called written description and utility re-
quirements.16 This latter group of guidelines presumably also had
some impact on the behavior of examiners in AU 1631.
This Article analyzes these dual influences in an early cohort of
patent applications assigned to AU 1631. It first compares the charac-
teristics of these applications with a comparison group of applications
from a related but more traditionally software-oriented data pro-
cessing art unit, AU 2123. Our analysis shows that on all conventional
measures of technological importance, private value and quality, ap-
plications in AU 1631 were significantly different from and “better”
than applications in AU 2123. To that extent, our results reinforce the
empirical theme of considerable variation in the manner in which the
system operates across technologies.
The Article then compares the examination of applications from
AU 1631 with a matched set of applications from AU 2123. On a
sample of applications matched on various dimensions of private val-
9. Id.
10. See USPTO, PATENT TECHNOLOGY CENTERS MANAGEMENT, http://www.uspto.gov/
web/patents/contacts/tcmgrs.htm (last visited Dec. 17, 2015).
11. Steinberg, supra note 7.
12. Id.
13. Id.
14. See generally AT&T Corp. v. Excel Commc’ns Inc., 172 F.3d 1352 (Fed. Cir. 1999);
State St. Bank & Tr. Co. v. Signature Fin. Grp. Inc., 149 F.3d 1368 (Fed. Cir. 1998).
15. Steinberg, supra note 7.
16. Revised Interim Guidelines for Examination of Patent Applications Under the 35
U.S.C. § 112, ¶ 1 “Written Description” Requirement; Request for Comments, 64 Fed. Reg.
71,427 (Dec. 21, 1999); Revised Utility Examination Guidelines; Request for Comments, 64
Fed. Reg. 71,440 (Dec. 21, 1999).
208 Harvard Journal of Law & Technology [Vol. 29
ue and quality, patent prosecution in AU 1631, with its biology-
trained examiners, looked strikingly different from prosecution in AU
2123. With the notable exception of nonobviousness rejections, appli-
cations in AU 1631 experienced more rejections, particularly notice-
related rejections, than a matched sample of applications in AU 2123.
As we discuss, these differences in patent examination quality,
particularly with respect to notice, appear to result from the biotech-
nology-specific examination guidelines that applied to examiners in
AU 1631, as well as the higher educational attainment of these exam-
iners. Our results therefore have implications for improving patent
examination quality. Specifically, our results suggest that the various
pressures that operate against rigorous examination, including in the
area of software, can be countered if properly trained examiners are
given notice-enhancing tools that can be applied with relative ease,
such as the written description and definiteness requirements.
We conclude by linking our results to the literature discussing
how the growing percentage of patent applications that rely on inter-
disciplinary and team-based science should be examined.
Part II of the Article summarizes the robust debate surrounding
patent quality and introduces the issue of measuring quality. Part III
surveys the quality metrics that scholars and policymakers have em-
ployed in evaluating the USPTO’s performance as an ex ante guaran-
tor of patent quality. Part IV applies the quality metrics identified in
Part III to more than five hundred patent applications from AU 1631
and AU 2123 and discusses the significant differences identified be-
tween applications reviewed in these two art units. Part V elaborates
on policy implications, both for patent quality and for how patent of-
fices should examine applications that increasingly rely on team-
based, interdisciplinary science.
II. QUALITY IN THE PATENT OFFICE
Much of the current dissatisfaction with the U.S. patent system
stems from concerns about patent quality. Numerous commentators
have put forward proposals for quality improvement.17 Some of the-
se — most notably, enhanced post-grant adjudications at the
USPTO — were incorporated into the America Invents Act of 2011
17. See generally JAMES BESSEN & MICHAEL J. MEURER, PATENT FAILURE (2008); DAN
L. BURK & MARK A. LEMLEY, THE PATENT CRISIS AND HOW THE COURTS CAN SOLVE IT
(2009); ADAM B. JAFFE & JOSH LERNER, INNOVATION AND ITS DISCONTENTS (2004); Stuart
Minor Benjamin & Arti K. Rai, Who’s Afraid of the APA? What the Patent System Can
Learn from Administrative Law, 95 GEO. L.J. 269 (2007).
No. 1] Bioinformatics at the Patent Office 209
(“AIA”).18 The court system has acted on others.19 Even so, more re-
mains to be done.
When critics bemoan poor quality, they are concerned about sev-
eral different issues.20 These include: failure to comply with the statu-
tory requirement that patents be granted only to inventions that would
not be obvious to the ordinary scientist or technologist working in the
area;21 failure to comply with the statutory requirement that the patent
disclose how to make and use the full scope of the invention covered
by the claims;22 and violation of the principle that the patent claims
must give proper notice as to the boundary of the patent right.23
Patent quality failures are perhaps most visible in litigation,24 due
to patent litigation’s high costs,25 often unpredictable outcomes,26 and
18. Leahy-Smith America Invents Act, Pub. L. No. 112–29, 125 Stat. 284 (2011) (codi-
fied in scattered sections of 35 U.S.C.).
19. For example, the Supreme Court in 2007 raised the nonobviousness standard for pa-
tentability by enabling patent examiners to combine prior art references and reject inven-
tions as obvious without strict application of the “teaching, suggestion, or motivation” test
to make such combinations. See KSR Int’l. Co. v. Teleflex Inc., 550 U.S. 398, 419 (2007).
20. For present purposes, we avoid the question of whether patents on certain types of in-
ventions are inherently “poor quality.” To be sure, the Supreme Court has clearly held that
restrictions should be placed on patent-eligible subject matter. See, e.g., Alice Corp. Pty. Ltd.
v. CLS Bank Int’l, 134 S. Ct. 2347, 2352 (2014); Ass’n for Molecular Pathology v. Myriad
Genetics, Inc., 133 S. Ct. 2107, 2111 (2013); Mayo Collaborative Servs. v. Prometheus
Labs., Inc., 132 S. Ct. 1289, 1294 (2012); Bilski v. Kappos, 561 U.S. 593, 611–12 (2010).
21. See 35 U.S.C. § 103 (2012); KSR, 550 U.S. at 420 (2007) (finding that a patent on an
adjustable automobile pedal assembly would, at the time of invention, have been obvious to
one of ordinary skill in the field when considered in light of the available technology).
22. See 35 U.S.C. § 112(a) (2012); Wyeth & Cordis Corp. v. Abbott Labs., 720 F.3d
1380, 1382 (Fed. Cir. 2009) (affirming that a patent on a method for using immunosuppres-
sant rapamycin drugs to treat and prevent re-narrowing of arteries required excessive exper-
imentation and was therefore invalid under § 112(a) for nonenablement).
23. See 35 U.S.C. §§ 112(b), (f) (2012); In re Packard, 751 F.3d 1307, 1309 (Fed. Cir.
2014) (affirming the USPTO’s rejection of a patent application on a coin change holder
under § 112(b) for indefiniteness); Power Integrations, Inc. v. Fairchild Semiconductor Int’l.,
Inc., 711 F.3d 1348, 1365 (Fed. Cir. 2013) (finding in pertinent part that patents on tech-
niques for mitigating electromagnetic interference and current flow problems recited suffi-
cient structure with respect to the term “soft start circuit” to satisfy the means-plus-function
requirements of § 112(f)). The Federal Circuit has also adopted the view that the written
description terminology of § 112(a) performs a notice function. Moreover, as discussed
further below, although the court indicated in 2010 that written description applies to all
technologies, during the time of this empirical study the requirement only appeared to apply
to biotechnology and chemistry. See Ariad Pharm. et al. v. Eli Lilly & Co., 598 F.3d 1336,
1349 (Fed. Cir. 2010) (en banc).
24. See, e.g., Charles Duhigg & Steve Lohr, Fighters in a Patent War, N.Y. TIMES (Oct.
7, 2012), http://www.nytimes.com/interactive/2012/10/08/business/Fighters-in-a-Patent-
War.html [http://perma.cc/23SS-ZV62] (summarizing a network of patent lawsuits among
the top patent litigants in the smart phone industry since 2006). Controversial litigation
arises not only in smartphones and software but also over genes. See, e.g., Brent Kendall,
Myriad Genetics Presses Ahead After High Court Ruling on Patents, WALL ST. J.: L. BLOG
(July 12, 2013), http://blogs.wsj.com/law/2013/07/12/myriad-genetics-presses-ahead-after-
high-court-ruling-on-patents/ [http://perma.cc/5YVK-B3VV].
25. Compare James Bessen & Michael J. Meurer, The Direct Costs from NPE Disputes,
99 CORNELL L. REV. 387, 390 (2014) (estimating the direct costs to defendants from patent
210 Harvard Journal of Law & Technology [Vol. 29
potentially lasting dynamic losses to innovation.27 Yet because many
of the more egregious harms from patent litigation are best understood
as effects rather than causes of poor patent quality,28 one important
locus of proposed reforms continues to be the USPTO infrastructure
for initial patent examination.29
Given the impossibility of systematically evaluating every patent
that issues from the USPTO for compliance with novelty, nonobvi-
ousness, disclosure, and notice requirements, critics have pointed to
various institutional features that would suggest poor quality. One set
of institutional features prevents proper application of existing patent
law standards. Another set may cause these standards to be too lax.
Because both sets of institutional features provide the motivation for
our empirical analysis of quality, we discuss them below. We then
discuss how these institutional implications vary in technology-
specific ways.
assertions by non-practicing entities at about $29 billion in 2011), with David L. Schwartz
& Jay P. Kesan, Analyzing the Role of Non-Practicing Entities in the Patent System, 99
CORNELL L. REV. 425, 433 (2014) (critiquing the findings and methodology of Professors
Bessen and Meurer as a likely upper bound and potentially biased upward).
26. Both the literature and case law identify a variety of sources for unpredictability in
patent litigation. See, e.g., J. Jonas Anderson & Peter S. Menell, Informal Deference: A
Historical, Empirical, and Normative Analysis of Patent Claim Construction, 108 NW. U. L.
REV. 1, 4–7 (2014) (discussing uncertainty in claim construction); Festo Corp. v. Shoketsu
Kinzoku Kogyo Kabushiki Co., 535 U.S. 722, 723 (2002) (explaining that despite the uncer-
tainty that the doctrine of infringement by equivalents introduces into the patent system, the
Court has repeatedly accepted such uncertainty “as the price of ensuring the appropriate
incentives for innovation”); see also Alan C. Marco & Saurabh Vishnubhakat, Certain
Patents, 16 YALE J.L. & TECH. 103, 132 (2013) (estimating the economic value of the cer-
tainty in patent rights that adjudicative resolution brings).
27. The subject of dynamic and static impacts on innovation from patenting is both wide-
ly studied in its own right and widely applied to particular economic and legal situations.
For a good overview, see generally Peter Lee, Toward a Distributive Commons in Patent
Law, 2009 WIS. L. REV. 917 (2009); Mark A. Lemley, The Economics of Improvement in
Intellectual Property Law, 75 TEX. L. REV. 989 (1997); Robert P. Merges & Richard R.
Nelson, On the Complex Economics of Patent Scope, 90 COLUM. L. REV. 839 (1990); Su-
zanne Scotchmer, Standing on the Shoulders of Giants: Cumulative Research and the Patent
Law, 5 J. ECON. PERSPS., Winter 1991, at 29.
28. Cf. Mark A. Lemley & A. Douglas Melamed, Missing the Forest for the Trolls, 113
COLUM. L. REV. 2117, 2120–21 (2013) (arguing that the effects of patent assertion by patent
trolls, however defined, are a symptom of systemic flaws in the patent system).
29. The situation post-grant appears to be quite different. Not only are the post-grant re-
view procedures set up by the AIA very heavily utilized, see AIA Trial Statistics, USPTO,
http://www.uspto.gov/patents-application-process/appealing-patent-decisions/statistics/aia-
trial-statistics (last visited Dec. 17, 2015) (tabulating usage of the AIA patent validity chal-
lenge procedures from April through September 2015), but criticisms center around the idea
that the procedures are too strict, not too lax. See, e.g., Ronald L. Grudziecki, Rapidly
Changing Patent Law Landscape Requires Careful Attention from Attorneys, ASPATORE,
May 2015 WL 3764843, at *3 (2015) (noting that the Patent Office “has received numerous
complaints regarding amendment proceedings, because some patents probably could have
been saved by further amendments, but the USPTO did not provide enough pages for
amendment, and the requirements were overly strict”).
No. 1] Bioinformatics at the Patent Office 211
A. Improper Application of Patent Law Standards
The USPTO operates under a number of institutional constraints
that might cause application of existing patent law standards to be too
lax. Perhaps most notably, examiners have a very limited amount of
time to examine patents.30 Thus, fact-intensive patent law standards
that involve significant work, such as thoroughly searching prior art to
evaluate nonobviousness or thoroughly evaluating an application’s
disclosure, may simply be too burdensome for examiners to imple-
ment successfully. Recent empirical work, discussed further in Part III,
suggests that these time constraints become particularly restrictive as
examiners advance in seniority.31
Another reason for concern is the agency’s funding structure,
which is not only entirely fee-based but also heavily based on fees
paid only if a patent is granted.32 The USPTO currently charges $1600
cumulatively for the filing, search, and examination of patent applica-
tions — even though the cost of doing this work is more than double
that amount.33 More generally, in the period from 2001 to 2014, over
half of the USPTO’s operating budget came from issuance fees that
are paid only after an examiner deems a patent application allowable
and from maintenance fees that are paid during the post-grant life of
the patent; filing, search, and examination fees from new applications
accounted for less than a third of the USPTO’s annual revenue.34 In
fact, some empirical research suggests that categories of patents from
which the USPTO is more likely to receive maintenance fees are also
30. See Michael D. Frakes & Melissa F. Wasserman, Is the Time Allocated to Review Pa-
tent Applications Inducing Examiners to Grant Invalid Patents?: Evidence from Micro-
Level Application Data 8–9 (Nat’l Bureau of Econ. Research Working Paper No. 20337,
2014), http://www.nber.org/papers/w20337 [http://perma.cc/BVC6-FG3R].
31. See generally id.
32. See, e.g., Arti K. Rai, Growing Pains in the Administrative State: The Patent Office’s
Troubled Quest for Managerial Control, 157 U. PA. L. REV. 2051, 2062 (2009) (arguing
that the USPTO’s fee structure “sets up an obvious financial incentive for the USPTO to
grant patents”).
33. The USPTO fee schedule for utility applications charges $280 for filing, $600 for
search, and $720 for examination, totaling $1600. USPTO Fee Schedule, USPTO (Jan. 1,
2014), http://www.uspto.gov/learning-and-resources/fees-and-payment/uspto-fee-schedule.
What these activities cost the USPTO to perform, however, is considerably higher. These
costs totaled $3569 in FY 2011, $3906 in FY 2010, and $3284 in FY 2009. USPTO,
USPTO SECTION 10 FEE SETTING — ACTIVITY-BASED INFORMATION AND COSTING
METHODOLOGY 18, http://www.uspto.gov/sites/default/files/aia_implementation/
aia_section_10_cost_supplement.pdf (last visited Dec. 17, 2015).
34. From FY 2001 through FY 2014, annual revenues from maintenance and issue fees
constituted 50% or more of total annual revenues, whereas annual revenues from filing,
search, and examination fees constituted between 24% and 31%. See USPTO Annual Re-
ports, USPTO, http://www.uspto.gov/about-us/performance-and-planning/uspto-annual-
reports (last visited Dec. 17, 2015) (containing hyperlinks to Performance and Accountabil-
ity Reports from FY 2001–2014).
212 Harvard Journal of Law & Technology [Vol. 29
more likely to be granted.35 In the absence of a specifically-identified
mechanism by which motivation for future economic gain might in-
fluence current USPTO behavior in particular patent areas, these find-
ings should be viewed cautiously. Nonetheless, given the USPTO’s
fee structure, applying patentability criteria loosely would have a
positive effect on USPTO funding. So would legal or policy decisions
that affirmatively relaxed these criteria. In the next section, we exam-
ine other institutional pressures that might cause legal standards to be
overly lax.
B. Lax Patentability Standards
Even absent substantive rulemaking authority, the USPTO has
some ability to articulate the contours of patent law.36 Beyond merely
exercising its discretion in the interstitial application of patent law to
the facts of patent examination, the USPTO also frequently issues
legal guidance documents to its examiners to interpret judicial deci-
sions and produces its own substantive precedents in the agency’s
internal administrative appeals process.37 Though the Federal Circuit
has, thus far, given no deference to these guidance documents or prec-
edents,38 the vanishingly small subset of USPTO actions that come
before the Federal Circuit reflects further limitations on the court’s
ability to actively manage the direction of substantive patent law.39
Moreover, as noted above, the USPTO’s funding structure may give it
an incentive to relax patentability standards.
Another mechanism by which patentability standards may have
become too lax involves the interaction between the USPTO and the
Court of Appeals for the Federal Circuit prior to the passage of the
America Invents Act of 2011. Prior to the AIA, the major route by
which appeals from the USPTO came to the Federal Circuit was
35. Michael D. Frakes & Melissa F. Wasserman, Does Agency Funding Affect Deci-
sionmaking?: An Empirical Assessment of the PTO’s Granting Patterns, 66 VAND. L. REV.
67, 69–71 (2013).
36. Melissa F. Wasserman, The PTO’s Asymmetric Incentives: Pressure to Expand Sub-
stantive Patent Law, 72 OHIO ST. L.J. 379, 387–89 (2011). See also Arti K. Rai, Engaging
Facts and Policy: A Multi-Institutional Approach to Patent System Reform, 103 COLUM. L.
REV. 1035, 1131–33 (2003) (comparing the relative competencies of the Federal Circuit in
judicial patent policymaking and the USPTO in administrative patent policymaking during
the early 2000s).
37. Wasserman, supra note 36, at 394–98.
38. Id. at 383. But see Arti K. Rai, Improving (Software) Patent Quality Through the Ad-
ministrative Process, 51 HOUS. L. REV. 503, 540–43 (2013) (arguing that legal determina-
tions made in post-grant review proceedings set up by the AIA may merit Chevron
deference).
39. Wasserman, supra note 36, at 398–400.
No. 1] Bioinformatics at the Patent Office 213
through a decision by an applicant to appeal a rejection.40 Thus, the
agency could be reversed only for improper rejections and not im-
proper grants. Under a model where the agency was motivated entire-
ly by the fear of Federal Circuit reversal, it would presumably reject
only that small subset of applications that were seen as unpatentable
even by the most “pro-patent” members of the Federal Circuit. The
result would be a progressive lowering of patent standards.
Although this model probably overstates the extent to which the
USPTO is motivated by Federal Circuit reversal, and understates the
extent to which it is influenced by the White House and by workload
fears, the USPTO has, at times, certainly found itself beholden to
Federal Circuit decisions that made it very difficult for the agency to
deny patents.41 For example, during the time period covered by our
study, the agency operated under a requirement that it show a written
teaching, suggestion, or motivation (“TSM”) to combine prior art
when making an obviousness rejection.42 As we discuss below, this
TSM requirement may have had particular force for AU 1631, where
prior art from the life sciences and software would presumably some-
times have to be combined to make an obviousness rejection.
C. Quality and Technology-Specificity
The debate over patent quality also has a technological dimension.
This dimension is particularly relevant for purposes of our paper, as
we explicitly address not simply a “recombinant” field but a field that
combines two areas — biotechnology and software — historically
considered quite different from a quality perspective.
Although the United States has a unitary patent system with few
formal exclusions or exceptions,43 the potentially wide-ranging differ-
ences among the economic and legal needs of various technology and
industry sectors have made it necessary and appropriate for U.S. pa-
tent law to adopt doctrines that accommodate those differences.44
Many substantive criteria for patentability operate by reference to
a person having ordinary skill in the art (“PHOSITA”), analogous to
40. See Jonathan Masur, Patent Inflation, 121 YALE L.J. 470, 503 (2011); Wasserman,
supra note 36, at 401–06.
41. See generally Arti K. Rai, Who’s Afraid of the Federal Circuit?, 121 YALE L.J.
ONLINE 335 (2011) (discussing the merits of Professor Masur’s inflationary model).
42. See id.; see also DyStar Textilfarben GmbH & Co. v. C.H. Patrick Co., 464 F.3d
1356 (Fed. Cir. 2006); In re Lee, 277 F.3d 1338 (Fed. Cir. 2002).
43. COMM. ON INTELL. PROP., NAT’L RESEARCH COUNCIL, A PATENT SYSTEM FOR THE
21ST CENTURY 45 (Stephen A. Merrill et al. eds., 2004).
44. Id.
214 Harvard Journal of Law & Technology [Vol. 29
tort law’s reasonably prudent person,45 and so incorporate technology-
specific perspectives into nominally technology-agnostic standards.46
The foundational role of the PHOSITA has, for example, resulted in a
doctrine of “unpredictable arts” that tolerates in those arts a greater
degree of experimentation for purposes of enablement47 and more
readily accepts unlikely advances over the prior art for purposes of
nonobviousness.48 Indeed, usage of the PHOSITA construct may itself
be so fact-intensive as to produce doctrine that is overly technology-
specific,49 raising the normative question of how best to identify the
PHOSITA.50
Beyond the PHOSITA construct, the Federal Circuit’s historical
tendency to apply the written description terminology of 35 U.S.C.
§ 112(a) to biotechnology and chemistry, but not necessarily to other
arts,51 had implications for both notice and scope. In the 1997 case of
University of California v. Eli Lilly, the Federal Circuit held that, even
for originally filed claims, written description was a requirement sepa-
45. Panduit Corp. v. Dennison Mfg., 810 F.2d 1561, 1566 (Fed. Cir. 1987) (noting patent
law’s reliance on “a ghost, i.e., ‘a person having ordinary skill in the art,’ not unlike the
‘reasonable man’ and other ghosts in the law”).
46. See 35 U.S.C. §§ 103, 112(a), 122(b)(2)(B)(v) (2012). See also Arti K. Rai, Building
a Better Innovation System: Combining Facially Neutral Patent Standards with Therapeu-
tics Regulation, 45 HOUS. L. REV. 1037, 1045 (2008).
47. See In re Wands, 858 F.2d 731, 736–37 (Fed. Cir. 1988).
48. See Procter & Gamble v. Teva Pharm. USA, Inc., 566 F.3d 989, 996 (Fed. Cir. 2009).
49. See generally Dan L. Burk & Mark A. Lemley, Is Patent Law Technology-Specific?,
17 BERKELEY TECH. L.J. 1155 (2002).
50. See, e.g., Jonathan J. Darrow, The Neglected Dimension of Patent Law’s PHOSITA
Standard, 23 HARV. J.L. & TECH. 227, 240–45 (2009) (arguing that the PHOSITA standard
has progressed from the tradesman “who practiced his art with ordinary skill but was not an
inventor” to the designer “whose work required a significant effort of the brain” to an inapt
researcher model that under-rewards innovation in the useful arts by defining the relevant
art itself in terms of innovative activity).
51. See, e.g., Moba v. Diamond Automation, Inc., 325 F.3d 1306 (Fed. Cir. 2003) (per
curiam), cert. denied, 540 U.S. 982 (2003). In Moba, Judge Rader wrote separately to note,
inter alia, that the written description requirement as interpreted in the Federal Circuit’s
jurisprudence had “create[d] a technology-specific rule in a technology-neutral statute.” Id.
at 1327 (Rader, J., concurring). Commentators have agreed with this assessment as well. See,
e.g., Burk & Lemley, supra note 49; Ajeet P. Pai, The Low Written Description Bar for
Software Inventions, 94 VA. L. REV. 457 (2008). In contrast, a 2007 paper by Chris Holman
uses data from appeals of patent rejections to the internal USPTO appeals board (then called
the Board of Patent Appeals and Interferences, or BPAI) to argue that written description
has not necessarily played an important role. Christopher M. Holman, Is Lilly Written De-
scription a Paper Tiger: A Comprehensive Assessment of the Impact of Eli Lilly and Its
Progeny in the Courts and PTO, 17 ALB. L.J. SCI. & TECH. 1 (2007). Holman emphasizes
the small number of BPAI decisions that address written description. However, to the extent
that written description is a relatively bright-line rule, a rejection that appears correct under
the application of that rule may not be appealed at all. E.g., Kate S. Gaudry & Joseph J.
Mallon, Appeals and RCEs — The Frequency and Success of Challenges to Specific Rejec-
tion Types, INTELL. PROP. TODAY, Nov. 2011, at 28 (finding that after a final rejection by
the examiner, applicant appeals to the BPAI as well as applicant requests for continued
examination relied quite infrequently on § 112 grounds such as written description or ena-
blement).
No. 1] Bioinformatics at the Patent Office 215
rate from enablement.52 The Federal Circuit also held that a genus
claim (in that case a functional genus claim to cDNA that coded for
insulin) satisfied written description only to the extent that it included
the structure of a subset of species representative of the genus.53 The
USPTO’s Written Description Guidelines, issued in draft form in
1999 and finalized in 2001,54 relied heavily on this precedent, stress-
ing that the structure of a “representative number of species” was nec-
essary to claim the genus.55
To the extent that the written description requirement has been
applied more to biotechnology and chemistry than to other arts,56 it
has arguably generated better boundary notice and more appropriately
tailored scope in biotechnology and chemistry. Boundary notice in
chemistry — though not necessarily in biotechnology — may also
reflect the discipline’s well-standardized conventions of nomencla-
ture.57 In contrast, the relatively imprecise vocabulary of software-
related inventions requires greater standardization through bodies,
such as the Institute of Electrical and Electronics Engineers
(“IEEE”).58
Additionally, because technology is itself malleable in definition,
imposing a priori classifications may compound the already difficult
task of comparing patent quality across different arts.59 One aspect of
this problem is that inventions that are truly seminal in a new field
52. 119 F.3d 1559, 1566 (Fed. Cir. 1997).
53. Id. at 1566–69 (invalidating functional genus claim to insulin cDNAs because the
written description failed to recite enough species to constitute a “substantial portion of the
genus”).
54. Guidelines for Examination of Patent Applications Under the 35 U.S.C. § 112, ¶ 1,
“Written Description” Requirement, 66 Fed. Reg. 1099, 1106 (Jan. 5, 2001).
55. Id.
56. See Margaret Sampson, The Evolution of the Enablement and Written Description
Requirements under 35 U.S.C. 112 in the Area of Biotechnology, 15 BERKELEY TECH. L.J.
1233, 1234–36 (2000). However, the Federal Circuit’s en banc decision in Ariad Pharm. v.
Eli Lilly & Co., handed down in 2010, does purport to make written description a require-
ment that applies equally to all technology. 598 F.3d 1336 (Fed. Cir. 2010) (en banc).
57. BESSEN & MEURER, PATENT FAILURE, supra note 17, at 152–53. But see generally
William D. Marsillo, How Chemical Nomenclature Confused the Courts, 6 U. BALT. INTELL.
PROP. L.J. 29 (1997) (arguing that where a genus of chemical compounds is to be claimed
by reference to a few representative species, the rules of chemical nomenclature have creat-
ed judicial confusion about theoretical permutations of chemical structure and the practical
import of actual chemical and physical properties).
58. See Peter S. Menell & Michael J. Meurer, Notice Failure and Notice Externalities, 5 J.
LEGAL ANALYSIS 1, 36 (2013) (comparing IEEE’s efforts in this regard to similar scientific
governance provided by the International Union of Pure and Applied Chemistry).
59. For example, what medical researchers may reliably have denoted “oncology” a cen-
tury ago is now an immensely broad collection of well-developed disciplines defined by
affected populations (e.g., pediatric oncology addressing cancer in children and geriatric
oncology addressing cancer in the elderly), affected biological systems (e.g., hematology-
oncology addressing blood-related cancers), and, in the case of personalized medicine, even
individual genomes. Modern science can no longer discuss inventions in the field of oncol-
ogy.
216 Harvard Journal of Law & Technology [Vol. 29
pose a challenge for measuring quality because, by their very nature,
they are quite broad in scope.60 Distinguishing broad patents from
overbroad patents is difficult.61 Evaluating the quality of such patents
is similarly difficult. Another aspect of the problem is that general-
purpose technologies, such as software, serve as platforms for, or in-
puts into, a wide variety of other fields,62 and inventions in these
widely adopted arts are not easily identified in objective and replica-
ble ways.63
This general debate over the importance of technology in patent
law informs a variety of specific patent quality discussions including
proposals for technology-agnostic64 and explicitly category-based65
reforms alike. Prominent among these proposals is the desire particu-
larly to assess the quality of software-related patents66 and to improve
it.67 These discussions raise a threshold boundary definition prob-
60. See Matthew J. Conigliaro, Andrew C. Greenberg & Mark A. Lemley, Foreseeability
in Patent Law, 16 BERK ELEY TECH. L.J. 1045, 1049–53 (2001) (describing pioneer inven-
tions as distinct from more incremental technological improvements and proposing a heuris-
tic for according due scope and protection to patents on pioneer inventions). But see
generally Brian J. Love, Interring the Pioneer Invention Doctrine, 90 N.C. L. REV. 379
(2012) (arguing from historical discussion that truly pioneering inventions do not exist and
that patent law should formally abrogate the doctrine of giving broad protection to pioneer
inventions).
61. See Apple Inc. v. Motorola, Inc., 757 F.3d 1286, 1334–38 (Fed. Cir. 2014) (Prost, J.,
concurring in part, dissenting in part) (discussing the need either to require adequate struc-
tural information in the patent or to apply means-plus-function treatment under 35 U.S.C.
§ 112(f), or else invite problems of overbroad patenting).
62. Stuart J.H. Graham & Maurizio Iacopetta, Nanotechnology and the Emergence of a
General Purpose Technology, 115/116 ANNALS OF ECONS. & STATS. 5, 8 (2014),
http://www.ssrn.com/abstract=1334376 [http://perma.cc/6S3N-AZU9]; Stuart J.H. Graham
& Saurabh Vishnubhakat, Of Smart Phone Wars and Software Patents, 27 J. ECON. PERSPS.
67, 74–75 (2013). See generally Timothy F. Bresnahan & Manuel Trajtenberg, General
Purpose Technologies: “Engines of Growth?” 65 J. ECONOMETRICS 83, 83–85 (1995)
(articulating the characteristics of general purpose technologies).
63. For a discussion of the methodological tradeoffs between reaching an accurate defini-
tion of software patents that minimizes Type I and Type II errors and reaching a precise
definition that is reproducible despite potential errors, see Bronwyn H. Hall & Megan Mac-
Garvie, The Private Value of Software Patents 13–17 (NBER Working Paper No. 12195,
May 2006), http://www.nber.org/papers/w12195 [http://perma.cc/5H8D-NBR8]. See also
Arti K. Rai, John R. Allison & Bhaven N. Sampat, University Software Ownership and
Litigation: A First Examination, 87 N.C. L. REV. 1519, 1526–33 (discussing Type I and
Type II errors in various definitions).
64. E.g., Jaffe & Lerner, supra note 17.
65. E.g., Peter S. Menell, A Method for Reforming the Patent System, 13 MICH.
TELECOMM. & TECH. L. REV. 487 (2007).
66. See John R. Allison, Abe Dunn & Ronald J. Mann, Software Patents, Incumbents,
and Entry, 85 TEX. L. REV. 1579, 1581 (2007) (framing proposed reforms for patents on
software-related inventions in terms of the “varying uses to which software firms put patents
in their businesses”); see also John R. Allison & Ronald J. Mann, The Disputed Quality of
Software Patents, 85 WASH. U. L. REV. 297 (2007).
67. Julie E. Cohen & Mark A. Lemley, Patent Scope and Innovation in the Software In-
dustry, 89 CAL. L. REV. 1, 3–4 (2001) (arguing for ex post tailoring of scope in software-
No. 1] Bioinformatics at the Patent Office 217
lem,68 which invites general reforms of patent quality that would dis-
proportionately affect patents on software-related inventions in a posi-
tive way,69 as well as software-specific reforms.70
More generally, quality concerns have spawned numerous pro-
posals for reform. Some of these proposals, such as significantly en-
hanced post-grant review of issued patents, have already been
implemented in the AIA. Others, particularly with respect to notice,
have yet to be attempted or have been implemented only in pilot form.
We defer discussion of future potential reforms to Part V, when we
turn to policy implications of our empirical findings.
In the next Part, we review efforts to measure patent quality quan-
titatively. Although these empirical studies do not necessarily lead
directly to specific normative conclusions, they provide important
background for our own empirical work.
III. MEASURING PATENT QUALITY
The challenge of describing patent quality in qualitative terms is
matched by the challenge of describing and estimating patent quality
through quantitative measures. Some of the difficulty arises because
the empirical literature has not always distinguished carefully between
at least three distinct visions of quality: (1) a patent document that is
“important” and facilitates diffusion of knowledge; (2) the private
value of a patent to the patent owner; and (3) a patent’s conformance
with existing legal criteria for patentability. 71 Additionally, an
emerging empirical literature views quality through the lens of exam-
iner characteristics and incentives that affect the rigor of examina-
tion.72
Below we review the quantitative measures in the existing litera-
ture, with a focus on those measures that will help us evaluate both
incoming patent applications to AU 1631 and AU 2123, and the ef-
related patents through a limited right to reverse-engineer as well as a narrow application of
the doctrine of equivalents).
68. Colleen V. Chien, Reforming Software Patents, 50 HOUS. L. REV. 325, 354–55 (2012)
(discussing the problem of defining software patents and noting some of the foundational
legal literature on proposed definitions).
69. See generally Rai, supra note 38, at 5–8.
70. Kevin Emerson Collins, Patent Law’s Functionality Malfunction and the Problem of
Overbroad, Functional Software Patents, 90 WASH. U. L. REV. 1399, 1400–02 (2013);
Mark A. Lemley, Software Patents and the Return of Functional Claiming, 2013 WISC. L.
REV. 905, 907–08 (2013).
71. See Ronald J. Mann & Marian Underweiser, A New Look at Patent Quality: Relating
Patent Prosecution to Validity, 9 J. EMPIRICAL L. STUD. 1, 2–4 (2012) (discussing literature
on knowledge diffusion and private value but focusing on legal validity); see also R. Polk
Wagner, Understanding Patent-Quality Mechanisms, 157 U. PA. L. REV. 2135, 2138–39
(2009) (distinguishing between patent quality and private patent value).
72. See infra Section III.D.
218 Harvard Journal of Law & Technology [Vol. 29
fects of examination. As we discuss in Part IV, we have data on the
characteristics of incoming applications and on how those applica-
tions were examined. At the art unit level, we also have data on a va-
riety of examiner characteristics.
A. Citations and Knowledge Transfer
The literature tabulating forward citations of patents73 stresses the
patent’s role in diffusing scientific or technical knowledge in a given
community. On this view, patents that are highly cited by other pa-
tents are likely to be important.74 One factor that is positively corre-
lated with forward citation rates, and with technological importance
more generally, is the number of co-inventors.75 For this reason, in
Part IV we use numbers of inventors as one metric for evaluating in-
coming patent applications.
73. See, e.g., Bronwyn H. Hall, Adam B. Jaffe & Manuel Trajtenberg, The NBER Patent
Citations Data File: Lessons, Insights and Methodological Tools (NBER Working Paper No.
8498, Oct. 2001), http://www.nber.org/papers/w8498.pdf [http://perma.cc/P4J7-9FDN];
Adam B. Jaffe, Manuel Trajtenberg & Rebecca Henderson, Geographic Localization of
Knowledge Spillovers as Evidenced by Patent Citations, 108 Q.J. ECON. 577, 577 (1993).
Numerous legal scholars have invoked this work in their own utilization of citation metrics.
See, e.g., James Malackowski & Jonathan A. Barney, What Is Patent Quality? A Merchant
Banc’s Perspective, 43 LES NOUVELLES 123, 130–31 (2008); Yasin Ozcan & Shane Green-
stein, Composition of Innovative Activity in ICT Equipment R&D, 45 LOY. U. CHI. L.J. 479,
493 (2013); Michael Risch, Patent Troll Myths, 42 SETON HALL L. REV. 457, 477–80
(2012).
74. See Malackowski & Barney, supra note 73, at 131–30. More formally, a higher de-
gree of citation (adjusted for time lag) should correspond with greater private value in the
patent for its owner. We note a recent empirical study that challenges this empirical assump-
tion. See David S. Abrams, Ufuk Akcigit & Jillian Popadak, Patent Value and Citations:
Creative Destruction or Strategic Disruption? (Penn Inst. Econ. Res., Working Paper No.
13-065, Nov. 2013), http://economics.sas.upenn.edu/system/files/13-065.pdf
[http://perma.cc/D5EH-MZB3] (finding that forward citation rates reflect growing private
value only initially, but later correspond to strategic behavior aimed at preserving the re-
turns from past patenting and to an overall decline in value). Though this study presents a
compelling early criticism of the citation-value literature, its applicability is limited in im-
portant respects. The underlying data is based on the patent portfolios of non-practicing
entities, whose reliance on licensing is a structurally different use of patent rights than by
practicing entities such as manufacturers, who gain more from actually excluding competi-
tors from the market than from merely threatening to exclude as leverage in a licensing
negotiation. The underlying data is also proprietary, making it difficult to reproduce or even
operationalize the findings into a more complete empirical model. This is not to suggest that
the findings themselves are incorrect, but that further, replicable research is needed before
the prevailing understanding of citations and value is properly discarded.
75. See Lee Fleming & Jasjit Singh, Lone Inventors as Sources of Breakthroughs: Myth
or Reality?, 56 MGMT. SCI. 41, 54 (2010) (arguing that a larger team increases the likeli-
hood of a breakthrough and decreases the likelihood of a relatively useless invention); Stef-
an Wuchty, Benjamin F. Jones & Brian Uzzi, The Increasing Dominance of Teams in
Production of Knowledge, 316 SCIENCE 1036, 1036–37 (2007).
No. 1] Bioinformatics at the Patent Office 219
The citation literature has also generated quality proxies, such as
technological originality and generality of the invention.76 Despite the
difficulty of separating out such general-purpose technology patents
for analysis,77 their presence in the background of so many different
technologies makes them measurable in their influence and impact.
Based on these insights, originality and generality have been dis-
cussed as correlate measures of patent quality.78
Recent empirical analysis has shown, however, that examiners
have a very significant influence on citations, and this influence is not
randomly distributed.79 Thus, it is unclear how citation data that pools
applicant and examiner citations should be interpreted. Indeed, one
study focused on variation among examiners found that examiners
whose patents were subsequently invalidated by the Federal Circuit
on average issued patents that were more frequently cited.80
For our purposes, because we focus on the quality of incoming
patent applications and of examination, the forward citation metric is
not relevant. On the other hand, as further discussed below, numbers
of backward citations to prior art perhaps intuitively address a basic
sense of quality.81 Particularly relevant for our purposes, applicant-
provided backward citations may be a proxy for the care with which
the applicant drafted the application. In Part IV, we use applicant cita-
tions to prior art (both patents and non-patent literature) to evaluate
incoming applications.
B. Private Value
The literature has often proxied for private value by looking at the
characteristics of patents that are litigated or for which renewal fees
are paid.82 In both cases, the assumption has been that rational parties
would not incur associated expenditures without some expectation of
76. See generally Hall et al., supra note 73 (deriving these measures from analysis of
time-adjusted citations).
77. In the context of software, the definition in most recent growing use relies on a de-
tailed sorting at the level of U.S. patent class and subclass to mitigate over- and under-
counting. See Graham & Vishnubhakat, supra note 62, at 75 n.7.
78. E.g., Josh Lerner, The Litigation of Financial Innovations, 53 J.L. & ECON. 807, 825
(2010).
79. Juan Alcácer, Michelle Gittelman & Bhaven Sampat, Applicant and Examiner Cita-
tions in U.S. Patents: An Overview and Analysis, 38 RES. POL’Y 415 (2009).
80. See Iain M. Cockburn, Samuel Kortum & Scott Stern, Are All Patent Examiners
Equal? The Impact of Characteristics on Patent Statistics and Litigation Outcomes, in
PATENTS IN THE KNOWLEDGE-BASED ECONOMY 19, 22 (Wesley M. Cohen & Stephen A.
Merrill eds., 2003).
81. Mann & Underweiser, supra note 71.
82. E.g., John R. Allison, Mark A. Lemley, Kimberly A. Moore & R. Derek Trunkey,
Valuable Patents, 92 GEO. L.J. 435, 437–38 (2004); Kimberly A. Moore, Worthless Patents,
20 BERKELEY TECH. L.J. 1521 (2005).
220 Harvard Journal of Law & Technology [Vol. 29
a return on investment. This literature has determined that litigated
and renewed patents generally have higher numbers of claims as well
as higher rates of forward and backward citation.83 Some commenta-
tors have therefore used number and complexity of claims not simply
as a reflection of private value but also as a proxy for effort expended
in obtaining the patent.84 Similarly, in Part IV, we use number of
claims at filing as a proxy for effort initially expended by the appli-
cant.85
Another important variable clearly correlated with private patent
value is the size of the patent family — that is, the number of foreign
jurisdictions in which the applicant has concurrently sought patent
protection for the same invention.86 We also rely on this variable in
Part IV.
C. Legal Validity
A third body of work examines the extent to which a particular
patent, examiner, or set of examination characteristics correlate with a
subsequent finding of legal validity. For purposes of what the USPTO
aims to do, the legal validity question is the most important. The va-
lidity metric is limited, however, by the very significant selection bias
involved in cases that are litigated to a final validity determination by
the Federal Circuit.87 The patent-level characteristics that predict
whether the small subset of patents litigated all the way to the Federal
Circuit will be found valid or invalid may not necessarily predict out-
comes with respect to the much larger pool of issued patents.88
83. Allison et al., supra note 82, at 438; Moore, supra note 82, at 1530.
84. E.g., David S. Abrams & R. Polk Wagner, Poisoning the Next Apple? The America
Invents Act and Individual Inventors, 65 STAN. L. REV. 517, 551 (2013).
85. Importantly, our use of the number of claims at filing is not a proxy for private value
per se. The private value literature considers the number of claims at issuance, and re-
striction requirements by examiners to divide an application’s claims into separate applica-
tions are particularly common in the fields of biotechnology and pharmaceuticals. As a
result, the number of claims at filing may not reflect private value because such a number
may differ from the number of claims at issuance. Yet the number of claims at filing does
reflect an applicant’s expectation, at the time of filing, of future private value in the pa-
tent — and so it is a reasonable proxy for applicant effort expended in producing a more
detailed, higher-quality application. See Abrams & Wagner, supra note 84, at 551 (includ-
ing the total number of claims in the empirical analysis, expecting “a higher-quality patent
to be more detailed, and thus have more claims”).
86. See, e.g., Dietmar Harhoff, Frederic M. Scherer & Katrin Vopel, Citations, Family
Size, Opposition and the Value of Patent Rights, 32 RES. POL’Y 1343 (2003).
87. See James Bessen & Michael J. Meurer, Lessons for Patent Policy from Empirical
Research on Patent Litigation, 9 LEWIS & CLARK L. REV. 1, 4 (2005).
88. That said, there is no reason to believe that the factors that predict validity or invalid-
ity in the subset of patents that reaches the Federal Circuit are strongly correlated with the
factors that determine review by the Federal Circuit in the first instance. Mann & Under-
weiser, supra note 71, at 22–23.
No. 1] Bioinformatics at the Patent Office 221
Bearing this caveat in mind, it is nonetheless worth noting that
one variable that is positively correlated with a finding of validity is
applicant-submitted prior art references.89 Indeed, the presence of ap-
plicant-submitted prior art references is a sufficiently prominent met-
ric that some scholars have thoroughly analyzed applicant behavior
across different technologies.90 The evidence indicates that applica-
tions in certain technology areas benefit from much more applicant-
supplied prior art than applications in other areas. For example, in a
sample of patents issued between January 1, 2001 and December 31,
2003, 45% of patents in the computers/communications and electri-
cal/electronic fields contained examiner-only citations compared to
only 25% of drug and medical patents and 30% of chemical patents.91
D. Examiner Characteristics
An emerging body of literature examines the interaction between
examiner characteristics and patent quality. Like the role of technolo-
gy, the role of examiners in patent quality is complex. Because many
aspects of an examiner’s work are technology-specific, the examiner
must at least be trained in a relevant science or engineering-related
discipline.92 As a consequence, we would expect to see at least some
variation in examiner characteristics. Indeed, as we discuss further in
Part IV, such variation appears to be a key factor in our results.
Beyond subject matter, however, institutional incentives also re-
sult in considerable variation in examiner behavior — variation that
has important implications for patent quality.93 For example, experi-
ence level could affect quality, and the correlation might be either
positive, because veteran examiners deliver more well-informed re-
views, or negative, because beginning examiners pay greater attention
to details than veterans. Experience also affects time allocated to ex-
aminers. Under a time allocation grid that has not been significantly
89. Mann & Underweiser, supra note 71, at 18.
90. Alcácer et al., supra note 79, at 419–20.
91. Id. The technological field effects were robust to nationality, assignee size, and other
factors. Id. For further detailed analysis of prior art supplied by applicants at the time of
filing, see generally Bhaven N. Sampat, When Do Applicants Search for Prior Art?, 53 J.L.
& ECON. 399 (2010).
92. Examiners in the mechanical engineering field, for example, must be proficient in
core subjects including differential and integral calculus, statics and dynamics, fluid me-
chanics and hydraulics, thermodynamics, electrical fields and circuits, properties and
strengths of materials, and optics. See Job Announcement No. CP-2014-0034, USAJOBS
(Mar. 21, 2014), http://www.usajobs.gov/GetJob/ViewDetails/365088500
[http://perma.cc/ZM6Z-HHE7].
93. See Cockburn et al., supra note 80.
222 Harvard Journal of Law & Technology [Vol. 29
revised since 1976,94 examiners at higher GS-levels within a given art
unit are allocated substantially less time to review applications than
examiners at lower GS-levels.95 Time pressure could exacerbate any
potential “burn out” felt by veterans.
Analysis at the individual examiner level is possible because,
within an art unit, applications appear to be randomly assigned.96 One
complication, however, is the reality that multiple examiners within
an art unit typically work on a single application. Primary examiners
who have authority to issue actions in their own name supervise and
sign off on the work of assistant examiners.97 Supervisory patent ex-
aminers oversee the work of entire art units98 in order to create more
uniformity in examiner performance.99
Despite this complication, some scholars have looked at individu-
al examiner characteristics. In one study that attributed patent exami-
nation to the assistant examiner when there was one, and to the prima-
primary examiner if there was no assistant examiner, Professors Mark
Lemley and Bhaven Sampat found that years of examiner experience
at the USPTO at the time of patent examination correlated inversely
with number of prior art references added and positively with grant
rate.100 More recently, Professors Michael Frakes and Melissa Was-
serman, studying individual examiners as they moved up the GS-level
ranks, found a similar trend in decreased examiner effort, and argued
that it emerges from unduly stringent time constraints placed by the
USPTO’s production quota system on higher-level examiners.101
94. Patent Examiner Count System, USPTO, http://www.uspto.gov/patent/
initiatives/patent-examiner-count-system (last visited Dec. 17, 2015) (noting that “the last
substantive change to the examiner goals was made over 30 years ago in 1976”).
95. Frakes & Wasserman, supra note 30. GS-level indicates General Schedule Pay Scale
for various government employees.
96. See Mark A. Lemley & Bhaven Sampat, Examiner Characteristics and Patent Office
Outcomes, 94 REV. ECON. & STAT. 817, 822 (2012).
97. U.S. Patent & Trademark Office, Manual of Patent Examining Procedure § 707.01
(9th ed. Rev. 7, Mar. 2014) [hereinafter MPEP].
98. See Patent Technology Centers Management, USPTO, http://www.uspto.gov/
patent/contact-patents/patent-technology-centers-management (last visited Dec. 17, 2015)
(listing supervisory patent examiners by art unit).
99. The effectiveness of these and related efforts at uniformity is the subject of ongoing
discussion and improvement. See, e.g., Todd J. Zinser, Memorandum for Director David J.
Kappos on the USPTO Patent Quality Assurance Process (Final Report No. OIG-11-006-I),
U.S. DEP’T OF COMMERCE OFFICE OF THE INSPECTOR GEN. (Nov. 5, 2010),
http://www.oig.doc.gov/OIGPublications/OIG-11-006-I.pdf [http://perma.cc/T9XY-WZ78]
(summarizing recommendations for improving the agency’s examination quality assurance
standards). Notably, although the input of several different examiners into patent examina-
tion is a challenge for studies that rely on the individual examiner as the unit of analysis, it
poses less of a challenge for our study, which takes the art unit as the unit of analysis.
100. Lemley & Sampat, supra note 96.
101. Frakes & Wasserman, supra note 30, at 3–5.
No. 1] Bioinformatics at the Patent Office 223
IV. BIOINFORMATICS AT THE USPTO: AN EMPIRICAL VIEW
This Part applies the quantitative patent quality literature dis-
cussed in Part III to an early cohort of patent applications from AU
1631 and a comparison group, AU 2123. We use the patent quality
measures identified above both to evaluate applications coming into
AU 1631 and AU 2123 and to isolate the effects of the examination
process on these applications.
A. Technology and Comparison Group
In general, AU 1631 broadly covers inventions combining biolo-
gy with computer implementation. The unit encompasses algorithms
that predict gene function and protein folding and the application of in
silico screening assays for identifying drug candidates.102
To understand more precisely what the USPTO does and does not
regard as bioinformatics technology, and why AU 2123 is a reasona-
ble comparison group, it is helpful to compare two classification sys-
tems that the USPTO employs. One is the United States Patent
Classification (“USPC”) system describing the technological fields to
which inventions pertain.103 The other is the USPTO Technology
Center system, which describes the organizational division of art units
that are responsible for patent examination.104 When the USPTO re-
ceives a patent application, the Office of Patent Classification both
categorizes it as to the technology class or classes that the claimed
invention best represents and assigns it to the art unit best suited to
examine it.105 Not surprisingly, there is a close concordance between
the USPC and the USPTO art unit hierarchy.106 According to this
102. The unit excludes biomedical imaging and simulation of organ functioning.
103. The USPC is not the only technology classification system maintained by the
USPTO — the International Patent Classification (“IPC”) system has long been in use as
well, and the Cooperative Patent Classification (“CPC”) system is the most recent initiative
for a harmonized taxonomy among the major patent systems. Classification Standards and
Development, USPTO, http://www.uspto.gov/patents/resources/classification (last visited
Dec. 17, 2015).
104. See generally Patent Technology Centers, USPTO, http://www.uspto.gov/
about/contacts/phone_directory/pat_tech (last visited Dec. 17, 2015). They are also referred
to as Patent Examiner Group Centers. See Office of the Deputy Commissioner for Patent
Operations, USPTO, http://www.uspto.gov/about-us/organizational-offices/office-
commissioner-patents/office-deputy-commissioner-patent (last visited Dec. 17, 2015).
105. See U.K. INTELL. PROP. OFFICE & U.S. PATENT & TRADEMARK OFFICE, PATENT
BACKLOGS, INVENTORIES AND PENDENCY: AN INTERNATIONAL FRAMEWORK 18 (2013),
http://www.ipo.gov.uk/ipresearch-uspatlog-201306.pdf [http://perma.cc/JUM7-XB2G]
(noting that applications received by the patent office first undergo pre-examination formali-
ties such as docketing and allocation according to relevant technology classification sys-
tems).
106. See Patent Classification: Classes Arranged by Art Unit, USPTO (Aug. 2015),
http://www.uspto.gov/patents-application-process/patent-search/understanding-patent-
224 Harvard Journal of Law & Technology [Vol. 29
concordance, bioinformatics inventions examined in AU 1631 are
classified into subsets of U.S. patent class 703 pertaining to data pro-
cessing.107 Traditional software informatics inventions examined in
AU 2123 are classified similarly.108 The two art units do not overlap
in the subclasses they cover. Between them, the art units cover class
703, as Table 1 summarizes.
Discussions with USPTO staff familiar with bioinformatics exam-
ination confirmed that AU 2123 is a closely complementary art unit to
AU 1631. USPTO staff noted, for example, that examiners in the two
art units sometimes share cases.109 In addition, because the technology
in each of the art units has been assigned the same class by the
USPTO, an examiner at a given GS-level in AU 1631 is given the
same amount of time to review an application as an examiner at the
same GS-level in AU 2123.110
Table 1: U.S. Patent Classes and Subclasses Mapped to
AUs 1631 and 2123
Class 703 (Data Processing: Structural Design, Modeling,
Simulation, and Emulation)
Subclass
Title
AU 1631
AU 2123
1
Structural Design
×
2
Modeling by Mathematical
Expression
×
3
Electrical Analog Simulator
×
4
Of Electrical Device or System
×
5
Of Physical Phenomenon
(e.g., Heat, Wave, Geophysics)
×
6
Simulating Nonelectrical Device
or System
×
7
Mechanical
×
8
Vehicle
×
classifications/patent-classification.html (providing a crosswalk between the two taxono-
mies).
107. Classes Arranged by Art Unit: Art Units 1611–1763, USPTO,
http://www.uspto.gov/patents-application-process/classes-arranged-art-unit-art-units-1611-
1763.html (last visited Dec. 17, 2015).
108. Classes Arranged by Art Unit: Art Units 1764–2691, USPTO,
http://www.uspto.gov/patents-application-process/classes-arranged-art-unit-art-units-1764-
2691.html (last visited Dec. 17, 2015).
109. Interview with Marjorie Moran, AU 1631 Supervisory Patent Examiner, and George
Elliott, former Technology Center 1600 Director (Oct. 19, 2013) (notes on file with author
Arti K. Rai) [hereinafter Interview with Moran and Elliot].
110. See Rai, supra note 32, at 2062–63 (discussing allotment of time by GS-level and
technology class).
No. 1] Bioinformatics at the Patent Office 225
9
Fluid
×
10
Well or Reservoir
×
11
Biological or Biochemical
×
12
Chemical
×
13
Simulating Electronic Device or
Electrical System
×
14
Circuit Simulation
×
15
Including Logic
×
16
Event-Driven
×
17
Event-Driven
×
18
Power System
×
19
Timing
×
20
Target Device
×
21
Computer or Peripheral Device
×
22
Software Program
(i.e., Performance Prediction)
×
23
Emulation
×
24
Of Peripheral Device
×
25
I/O Adapter
(e.g., Port, Controller)
×
26
Of Instruction
×
27
Compatibility Emulation
×
28
In-Circuit Emulator (I.E., ICE)
×
However, though the technology in the applications allocated to
the two art units is comparable, applications received in the two art
units may not necessarily be comparable. As noted previously, drugs,
medicine, and chemistry all represent areas where applicants have
historically supplied significantly more prior art than in electronics
and communications.111 Whether or not bioinformatics applications
resembled drugs, medicine, and chemistry in this respect was a propo-
sition we tested.
In addition, we tested for differences in patent examination based
on examiner characteristics at the level of the art unit. On average,
patent examiners in the two art units differ in several ways that may
be meaningful. Given AU 1631’s biological sub-focus within infor-
matics, examiners in that art unit are primarily trained in a biological
science with additional relevant expertise in computer science, rather
than primarily trained in computer science as AU 2123 examiners
are.112 USPTO staff also indicated that AU 1631 has more examiners
111. See supra Part III.C.
112. Interview with Moran and Elliot, supra note 109.
226 Harvard Journal of Law & Technology [Vol. 29
who hold advanced degrees than AU 2123: approximately 55% to
60% of AU 1631 examiners have Ph.D. degrees and up to 90% have
master’s degrees.113
In this regard, personnel data obtained through Freedom of In-
formation Act (“FOIA”) requests, and generously provided to us by
Professor Ronald Mann, showed differences in the personnel in the
respective art units. Specifically, the thirteen examiners in AU 1631 in
calendar year 2003 had a median GS-level of 13 and had been at the
USPTO for a median of four years.114 The sixteen examiners in AU
2123 had a median GS-level of 11.5 and had been at the USPTO for a
median of two years.115
Finally, for reasons introduced in Part II.C and discussed further
below, the particular legal details of examination may have been quite
technology-specific. This was true not only because of practical dif-
ferences in the definition of a PHOSITA116 and the relative unpredict-
ability of the art,117 but also because of mechanisms by which USPTO
guidelines and Federal Circuit case law, particularly in 2003, may
have applied differentially to bioinformatics and “ordinary” soft-
ware.118
In all, we were interested both in potentially divergent character-
istics of applications as they entered the two art units, and divergent
treatment that awaited them there. As to the latter question, even as-
suming comparable inputs, the literature led us to form several hy-
potheses about differential examination.
B. Hypotheses About Examination
First, we expected significant differences in treatment under the
written description requirement of 35 U.S.C. § 112(a). Not long after
the creation of AU 1631, the USPTO in early 2001 finalized its
Guidelines for Examination of Patent Applications Under the 35
U.S.C. § 112 ¶ 1, “Written Description” Requirement.119 Although
the guidelines did not, on their face, apply only to biotechnology, all
of their examples came from biotechnology, and most of the relevant
113. Id.
114. FOIA-Requested USPTO Personnel Data (on file with author Arti K. Rai).
115. Id.
116. See supra notes 45–50.
117. In re Fisher, 427 F.2d 833, 839 (C.C.P.A. 1970) (explaining that the scope of ac-
ceptable instruction in the patent varies inversely with the level of scientific or technological
unpredictability that is involved).
118. Supra notes 46–49 and accompanying text (discussing divergent standards of suffi-
ciency as to patentability based on the unpredictability of the field).
119. 66 Fed. Reg. 1099 (Jan. 5, 2001).
No. 1] Bioinformatics at the Patent Office 227
Federal Circuit case law at the time addressed biotechnology.120 Thus,
we expected more stringent application of the written description re-
quirement to applications assigned to AU 1631. We further hypothe-
sized that examiners would find it difficult to confine written
description rejection to the strictly biological aspects of applications,
and thus this stringent application would “spill over” into the “soft-
ware” aspects of the invention — the net result being that applications
would sometimes be rejected for failing to fully describe relevant al-
gorithms and data.
Notably, we expected stricter application of the written descrip-
tion requirement, even though the literature on examiner characteris-
tics discussed in Part III might lead us to believe that the higher GS-
level examiners in AU 1631 would be less likely to administer a thor-
ough examination. Our prediction was that higher education levels
and clear expectations that written description should apply strictly to
applications involving biotechnology would counteract any effects
arising from examiner GS-level.
Second, we wanted to examine any differences in other aspects of
the patent law that implicate notice — specifically: definiteness, dou-
ble patenting, and restriction requirements. Here, we did not have a
specific hypothesis but were motivated by the persistent criticism that
software patents fail in their notice function.
Third, we expected examination in AU 1631 to more strictly ap-
ply the utility requirement of 35 U.S.C. § 101. At the same time it
issued its Written Description Guidelines, the USPTO issued its Utili-
ty Examination Guidelines.121 Although these guidelines were nomi-
nally agnostic as to technology, they were written with biotechnology
in mind.122 Thus, we expected more stringent application of the utility
requirement to applications that claimed biotechnology-related inven-
tions. Again, our prediction was that higher education levels and clear
expectations of how to examine applications touching on biotechnol-
ogy would counteract any effects arising from GS-level.
Fourth, we expected differences in treatment under both the ena-
blement requirement of 35 U.S.C. § 112(a) and the nonobviousness
requirement of 35 U.S.C. § 103. With respect to enablement, we pre-
dicted that the greater unpredictability of biotechnological arts and the
impact of such unpredictability on what a person of ordinary skill in
the art would consider enabled would lead to a higher rejection rate.
Again, our prediction was that higher education levels and clear ex-
120. See Arti K. Rai, Patent Validity Across the Executive Branch: Ex Ante Foundations
for Policy Development, 61 DUKE L.J. 1237 (2012) (discussing the tortured history of the
written description and utility guidelines).
121. 66 Fed. Reg. 1092 (Jan. 5, 2001).
122. See Rai, supra note 120.
228 Harvard Journal of Law & Technology [Vol. 29
pectations of how to examine applications touching on biotechnology
would counteract any effects of GS-level.
As for nonobviousness, we expected that the strict requirement
that existed prior to 2007 for an examiner to identify teaching, sugges-
tion, or motivation in order to combine prior art123 would lead to a
lower nonobviousness rejection rate for bioinformatics because of its
interdisciplinary nature. This effect would arise on top of any differ-
ence in nonobviousness rejection rate caused by unpredictability in
the field. In this case, GS-level effects might be a third factor leading
to lower rates of nonobviousness rejection.
C. Results: Applications
To test our hypotheses and describe more fully the contours of
USPTO bioinformatics examination early in its history, we reviewed
by hand the prosecution histories of patent examinations that were
filed between January 1 and December 31, 2003, and were assigned to
AU 1631 or AU 2123. These examination records are publicly availa-
ble from the USPTO Patent Application Information Retrieval (PAIR)
system124 as well as in bulk through Google125 and, more recently,
Reed Tech.126
We chose calendar year 2003 because it represented the first year
for which full prosecution history data on all applications were avail-
able on PAIR. Additionally, by 2003, AU 1631 had become reasona-
bly well-established. However, at the same time, the application of
software to biology was not yet considered entirely routine. Thus, for
example, in 2003, bioinformatician Lincoln Stein famously gave bio-
informatics “ten years to live,”127 predicting that although informatics
would continue to expand and be used, it would, within ten years, be
absorbed into biology.128
For each prosecution, we collected information on patent applica-
tion characteristics, which, as the empirical literature discussed in Part
III suggests, bears on one or more of the following: (1) scientific im-
portance; (2) private value; and (3) legal validity. Specifically, we
123. This requirement was relaxed in KSR Int’l Co. v. Teleflex Inc. 550 U.S. 398, 401
(2007).
124. Patent Application Information Retrieval, USPTO,
http://portal.uspto.gov/pair/PublicPair (last visited Dec. 17, 2015).
125. USPTO Bulk Downloads: Patents, GOOGLE,
http://www.google.com/googlebooks/uspto-patents.html [http://perma.cc/G6E2-KRP4].
126. USPTO Data Sets, REED TECH. & SVCS. INC., http://patents.reedtech.com/patent-
products.php [http://perma.cc/HS33-MAEL].
127. Daniel H. Steinberg, Stein Gives Bioinformatics Ten Years to Live, O’REILLY
MEDIA (Feb. 5, 2003), http://archive.oreilly.com/pub/a/network/biocon2003/stein.html
[http://perma.cc/N397-DFWW].
128. Id.
No. 1] Bioinformatics at the Patent Office 229
collected information on: (1) number of inventors (scientific im-
portance); (2) number of claims; (3) family size; and (4) application
and/or grant in the European Patent Office (“EPO”) and Japan Patent
Office (“JPO”) (private value) and applicant-cited prior art, including
Non-Patent Literature (“NPL”) (legal validity).
In total, we gathered data on 565 prosecution histories, 393 from
AU 1631 and 172 from AU 2123. Univariate analysis was performed
using unpaired two-tailed t-test for continuous variables and Fisher’s
exact test for categorical variables. Significance was assessed using an
alpha of 0.05. Table 2 illustrates the differences in these measures
among applications in both art units. Virtually all the differences are
statistically significant at the mean, as Table 2 shows.
Table 2: Two-Tailed Comparison of Means for Application and
Examination Process Characteristics
Characteristic
Mean
(AU 1631)
Mean
(AU 2123)
Inventors ***
3.36
2.31
Claims ***
38.41
26.15
Applicant Cited Prior Art ***
30.17
8.72
Applicant Cited NPL ***
20.97
4.41
Family Size ***
3.05
2.26
Applied in EPO ***
0.48
0.26
Granted in EPO *
0.12
0.06
Applied in JPO
0.35
0.30
Granted in JPO *
0.09
0.17
Family Size ***
3.05
2.26
*p < 0.05; **p < 0.01; ***p < 0.001
Thus, notwithstanding the fact that both sets of applications in-
volved software used for modeling and simulation, applications in AU
1631 were more scientifically important (number of inventors), pri-
vately valuable (numbers of claims, family size, and application in the
EPO and JPO), and likely to be legally valid (applicant cited prior art,
including NPL), than applications in AU 2123.
To some extent, our results on incoming patent applications are in
line with, and reinforce, the theme of quality variation across patent-
ing in different technologies that we discussed in Parts II.C and III.C.
That said, the fact that such variation occurs even when the actual
scientific difference between the technologies in question (software
for biological modeling vs. general modeling software) is quite mod-
est is striking.
230 Harvard Journal of Law & Technology [Vol. 29
D. Results: Examination
Although the disparity in application characteristics was not
unexpected, it did indicate to us that we needed to create a new data
set for purposes of testing examination results. Specifically, we decid-
ed to create a subset that consisted of a matched sample of applica-
tions in the two art units. We matched applications on the following
variables:
• Number of inventors at filing;
• Number of claims at filing;
• Number of applicant-cited prior art references;
• Number of applicant-cited non-patent literature
references;
• Patent family size;
• Whether concurrent application in the EPO was
sought;
• Whether concurrent application in the EPO was
granted;
• Whether concurrent application in the JPO was
sought; and
• Whether concurrent application in the JPO was
granted.
As to the latter four variables (pertaining to concurrent applica-
tions), pairs were matched directly. That is, an application from AU
1631 was required to have the same value (true vs. false) as the poten-
tial paired application from AU 2123. For the former five variables,
pairs were quartile-matched. For example, an application from AU
1631 may have been in the first quartile of the distribution of number
of inventors at filing for all AU 1631 applications. Such an applica-
tion could be paired only with an application from AU 2123 that was
also in the first quartile of the distribution of number of inventors at
filing for all AU 2123 applications. Where multiple potential pairs
existed that satisfied all matching criteria, pairs were assigned ran-
domly.129 This matched sample consisted of sixty-one applications in
each art unit. The relatively small size of this matched sample may
cause certain examination disparities between the samples to fail to
reach statistical significance. Thus, to the extent we discuss only dis-
parities that have statistical significance, we probably underestimate
disparities.
129. We built the sets of potential pairs by hand in Excel and used that software’s ran-
dom number generator to assign pairs randomly.
No. 1] Bioinformatics at the Patent Office 231
With respect to the substance of examination, we collected infor-
mation on the statutory grounds for rejections, the source of the prior
art that the examiner cited in support of such rejections, and the exam-
iner’s imposition of a restriction requirement, if any, through the first
round of examination.130 Tables 3 and 4 illustrate our findings.
As these Tables show, examination in the two art units differed
quite significantly. We discuss these differences in detail in the next
section.
Table 3: Grounds for Rejection During Prosecution
Grounds
Proportion
(AU 1631)
n = 393
Proportion
(AU 2123)
n = 172
p-value
Subject-Matter
Eligibility
40.97%
43.02%
0.6481
Utility
5.60%
0.58%
0.0055
Novelty
59.29%
55.81%
0.4411
Nonobviousness
52.67%
80.81%
0.0000
Enablement
24.68%
8.14%
0.0000
Definiteness
66.41%
54.65%
0.0078
Written Description
25.19%
8.14%
0.0000
Double Patenting
16.28%
5.81%
0.0007
Restriction
79.90%
8.72%
0.0000
Table 4: Source of Prior Art Used in Rejection131
Source
Proportion
(AU 1631)
n = 393
Proportion
(AU 2123)
n = 172
p-value
Prior Art Cited by the
Examiner
44.53%
66.86%
0.0000
Prior Art Cited by the
Applicant
7.38%
3.49%
0.0075
Prior Art Cited by Both
17.56%
27.91%
0.0052
130. A restriction requirement is an enforcement by the examiner of the single-invention
rule and consists of a finding by an examiner that a patent application claims “two or more
independent and distinct inventions,” so that the applicant must restrict the application to
one of them. 35 U.S.C. § 121 (2012). The applicant may then claim each remaining inven-
tion in a divisional application that otherwise satisfies the criteria for patentability. See
generally 37 C.F.R. § 1.142 (2015); MPEP Ch. 0800 (9th ed. Rev. 7, Mar. 2014).
131. The values for each art unit do not add up to 100% because there were a number of
applications in both samples (an especially large number in AU 1631) for which prior art
was not a basis for rejection.
232 Harvard Journal of Law & Technology [Vol. 29
E. Discussion
1. Subject Matter
The rate at which bioinformatics applications in AU 1631 re-
ceived rejections for subject matter ineligibility under § 101 (36.87%)
was not significantly different from that of software applications in
AU 2123 (42.62%).132 As we have discussed, given that bioinformat-
ics and other informatics bear a general taxonomic similarity and dif-
fer only in their details.133 Thus, it is perhaps unsurprising that the
requirements of subject matter did not affect one group more than the
other.
That said, subject matter is a rather coarse filter for evaluating in-
ventions. Indeed, this view appears in the case law,134 the literature,135
and even legal guidance that the USPTO has issued in response to the
Supreme Court’s interpretation of § 101.136 For a more fine-grained
comparison, we turn to various patentability requirements under § 112
and § 103.
2. Written Description: Impacts on Notice
Supporting our first hypothesis, bioinformatics applications in
AU 1631 did receive rejections for inadequate written description un-
der § 112 to a significantly greater extent (21.30%) than did software
applications in AU 2123 (6.56%).137 Moreover, as we predicted, use
of written description was not limited to the biological aspects of the
invention; a check of the first fifteen written description rejections in
AU 1631 for which we coded showed that thirteen of the fifteen rejec-
tions involved examiner arguments that the applicants had failed to
describe adequately either an algorithm or relevant data.138
132. See supra Table 3 (finding no statistically significant difference at the 95% confi-
dence interval). Similarly, for novelty, the difference between rejections received for appli-
cations in AU 1631 (49.18%) and AU 2123 (55.74%), was not statistically significant. See
id.
133. See supra Part IV.A.
134. E.g., Res. Corp. Tech., Inc. v. Microsoft Corp., 627 F.3d 859, 869 (Fed. Cir. 2010)
(referring to § 101 as “the coarse eligibility filter”).
135. E.g., Michael W. Carroll, One for All: The Problem of Uniformity Cost in Intellec-
tual Property Law, 55 AM. U. L. REV. 845, 892–93 (observing that “use of standards along
the subject matter dimension permits only coarse-grained exercise of interpretive discretion
because an adjudicator can choose only between applying all or no rights to a particular
innovation or class of innovations”).
136. Interim Guidance for Determining Subject Matter Eligibility for Process Claims in
View of Bilski v. Kappos, 75 FED. REG. 43,922, 43,926 (July 27, 2010) (explaining that
“Section 101 is merely a coarse filter”).
137. Id.
138. The written description rejections we examined arose in application numbers
10/204849, 10/304496, 10/309152, 10/309391, 10/332999, 10/345905, 10/350341,
No. 1] Bioinformatics at the Patent Office 233
As previously discussed, adequate written description requires
disclosure of structure commensurate with the scope of what is
claimed.139 As a result, it may somewhat limit scope, perhaps unduly.
The written description requirement’s chief virtue is its promotion of
boundary notice. In turn, certainty about patent boundaries offers a
number of benefits including the ability to assess the value of patent
rights for transaction and commercialization,140 to “distinguish the
invention or discovery from other things before known and used”141 in
determining freedom to operate, and, most simply, to avoid infringe-
ment of a competitor’s patents.142
Taken in historical context, this finding as to bioinformatics in-
ventions in AU 1631 and more conventional informatics inventions in
AU 2123 suggests that the structural specificity that already existed
for biological and biochemical inventions invited more robust exam-
iner scrutiny of the relatively unstructured software elements of appli-
cations in AU 1631. The alternative inference is that applications on
conventional software informatics inventions in AU 2123 received
fewer written description rejections, because they were already better
described under § 112. This inference, however, is inconsistent with
the widely accepted view, discussed in Part II.C, that conventional
software patents pose substantial challenges for notice. Moreover, as
we discuss further in Part V, current patent notice reform efforts
aimed at improving the correlation between functional claiming and
corresponding structure also focus heavily on software.
3. Other Indicia of Notice
Our results also showed differences in examination with respect
to other indicia of patent notice. Three measures in our results — def-
initeness, double patenting, and restriction — were particularly rele-
vant in this regard. The definiteness requirement promotes boundary
notice by requiring that patent claim terms clearly delineate ex ante
10352246, 10/359439, 10/360747, 10/360796, 10/363727, 10/432932, 10/430685, and
10/378866. Of these, only two (10/304496 and 10/359439) applied written description to the
biological aspect of the invention.
139. See supra note 22 and accompanying text.
140. See generally Craig Allen Nard, Certainty, Fence Building, and the Useful Arts, 74
IND. L.J. 759 (1999). In advocating for opposition proceedings to test closely the validity of
patents and provide early certainty, Professor Nard argues that such a proceeding “will
facilitate greater accuracy in private valuation because, as the prior art picture becomes
more complete during prosecution, the more informed the parties will be with respect to the
boundaries of the claimed invention.” Id. at 765–66.
141. Nautilus, Inc. v. Biosig Instruments, Inc., 134 S. Ct. 2120, 2124–25 (2014) (citing
the first Patent Act, Act of Apr. 10, 1790, § 2, 1 Stat. 110; the Act’s call for definiteness in
claims continues to animate patent law to the present day).
142. See Halliburton Energy Servs., Inc. v. M-I LLC, 514 F.3d 1244, 1249 (Fed. Cir.
2008); Laitram Corp. v. Cambridge Wire Cloth Co., 863 F.2d 855, 856–57 (Fed. Cir. 1988).
234 Harvard Journal of Law & Technology [Vol. 29
the metes and bounds of an invention as claimed in the patent.143 The
double patenting rule furthers this goal by exerting downward pres-
sure on the sheer quantity of rights that a market actor must navigate
and clear, particularly because the existence of duplicative patent
rights injures the public’s expectation that the expiration of a patent
on an invention will leave that invention free for public use.144 This
rule recognizes that even if individual patent rights were perfectly
clear ex ante — an assumption that is far from realistic — the discov-
ery costs of identifying relevant rights145 are also a function of scale.
Put another way, if the capacity to search effectively does not survive
increasing size and complexity in the set of all patent rights, then
search bottlenecks may still cause notice failures.146
Unlike definiteness and double patenting, which clearly promote
notice, the single invention rule has more complex effects. On the one
hand, it furthers notice by mitigating the complexity of patent rights in
somewhat the same way that the definiteness requirement seeks to do.
Whereas definiteness offers clarity in evaluating claims within a pa-
tent, restriction offers clarity in evaluating inventions within a patent.
Examiners commonly impose restriction requirements in pharmaceu-
tical-and biotechnology-related applications because it is common for
applications in those fields to claim numerous related chemical com-
pounds or processes that turn out to be patentably distinct.147
That said, there is evidence that divisional applications occur not
only among the least valuable patents, where applicant ignorance
might be the cause, but also among the most valuable patents, sug-
gesting that sophisticated applicants sometimes draft claims calculated
to provoke restriction by the examiner.148 This evidence may mean
that applicants who draft claims calculated to provoke restriction
sometimes intend to produce strategic delay in examination, and that
143. See All Dental Prodx, LLC v. Advantage Dental Prods., Inc., 309 F.3d 774, 779–80
(Fed. Cir. 2002) (citing Warner-Jenkinson Co. v. Hilton Davis Chem. Co., 520 U.S. 17, 28–
29 (1997)).
144. In re Longi, 759 F.2d 887, 892–93 (Fed. Cir. 1985) (explaining that “[t]he public
should . . . be able to act on the assumption that upon the expiration of the patent it will be
free to use not only the invention claimed in the patent but also [obvious] modifications or
variants”).
145. Discovery costs are one subset of the general transaction costs that Ronald Coase
distinguished in his theory of externalities. See R. H. Coase, The Problem of Social Cost, 3
J.L. & ECON. 1, 15 (1960).
146. For a systematic overview of this principle, see generally Christina Mulligan &
Timothy B. Lee, Scaling the Patent System, 68 N.Y.U. ANN. SURV. AM. L. 289 (2012).
147. Matt Browning, Now You See Them, Now You Don’t: The USPTO’s Rules on
Claims and Continuations, 23 BERKELEY TECH. L.J. 247, 251 (2008).
148. See Stuart J. H. Graham & David C. Mowrey, Submarines in Software? Continua-
tions in US Software Patenting in the 1980s and 1990s, 13 ECON. INNOVATION & NEW
TECH. 443 (2004).
No. 1] Bioinformatics at the Patent Office 235
this practice is particularly common in biotechnology.149 If so, re-
striction may actually hinder public notice by creating market uncer-
tainty about whether later-issued patent rights may encumber sunk
investments.150 In any case, the paradox of the restriction require-
ment’s notice benefit is that, by carving up multi-invention applica-
tions into patentably distinct sets of rights, the single invention rule
actually increases the total set of rights to be searched and cleared,151
but it does so by reducing the patent-to-invention relationship to a
simple one-to-one correspondence.152
With respect to all three rejection grounds in our matched sample,
the percentage of rejections in AU 1631 was higher. And on two of
the three rejection grounds, the difference was statistically significant
at the power of the study. With respect to indefiniteness, applications
in AU 1631 received more rejections (62.30%) than did applications
in AU 2123 (40.98%).153 Second, with respect to the double-patenting
rule, applications in AU 1631 received more than twice as many re-
jections (13.11%) than did applications in AU 2123 (4.92%),154 alt-
hough this second result was not statistically significant at the power
of this study. Third, with respect to the single invention requirement
of § 121 to manage the complexity of patent rights by limiting each
patent to one invention,155 applications in AU 1631 received an order
of magnitude more rejections (70.49%) than did applications in AU
2123 (6.56%).156
Unlike with written description, no specific USPTO guidance in-
structed examiners in AU 1631 to apply these other notice require-
ments vigorously. The reason for generally higher levels of rejection
in AU 1631 is therefore not clear. It is possible that higher educational
levels could have led examiners in AU 1631 to police notice more
149. Mark A. Lemley & Kimberly A. Moore, Ending Abuse of Patent Continuations, 84
B.U. L. REV. 63, 103 n.164 (2004).
150. Id. at 72–73 (discussing the relationship between examination delay and market un-
certainty). Professors Lemley and Moore argue further that 18-month publication of pending
applications after the 1999 AIPA may not meaningfully have improved transparency about
pending applications, as publication is required only for applications that will also be filed
abroad, where 18-month publication is the norm, meaning that U.S. publication reveals only
what would have become publicly available regardless. Id. at 88–89.
151. Supra notes 145–46 and accompanying text.
152. Notably, the professed purpose of restriction practice is usually administrative con-
venience in patent examination. See, e.g., Applied Materials, Inc. v. Advanced Semiconduc-
tor Materials Am., Inc., 98 F.3d 1563, 1568–69 (Fed. Cir. 1996). The extent to which
restriction practice actually strikes an efficient balance in patent notice is an open empirical
question.
153. See supra Table 3.
154. See supra Table 3.
155. See supra note 130.
156. See supra Table 3.
236 Harvard Journal of Law & Technology [Vol. 29
vigilantly, despite the fact that examiners in AU 1631 were more
time-constrained on average than examiners in AU 2123.157
Supporting our third hypothesis, bioinformatics applications in
AU 1631 received rejections for inadequate utility under § 101 much
more frequently (8.20%) than the traditional software informatics ap-
plications in AU 2123 (0.00%).158 Possibly the Utility Guidelines that
the USPTO issued in 2001, requiring a specific assertion by the appli-
cant of the utility of the claimed invention,159 had some effect on ex-
aminers.
4. Enablement and Nonobviousness
The results did not support our hypothesis that bioinformatics ap-
plications in AU 1631 would receive significantly more rejections for
inadequate enablement under § 112 (a) than would traditional soft-
ware informatics applications in AU 2123.160 Arguably, the most no-
table feature of the enablement results was the low frequency of these
rejections across both art units. This low frequency is consistent with
the view that enablement rejections are complex, fact-and prior art-
intensive inquiries that time-and resource-constrained examiners are
unlikely to favor.
Consistent with our hypothesis, applications in AU 2123 did re-
ceive significantly more rejections for obviousness (78.69%) than did
applications in AU 1631 (39.34%).161 In general, across all three
hundred ninety-nine AU 1631 prosecution histories that we analyzed,
examiners conducted prior art searches in both biology and software
in the majority of cases. However, given that bioinformatics was still
an emerging interdisciplinary field as of 2003, and given that the rele-
vant pre-KSR law at the time required a very specific teaching, sug-
gestion, or motivation in the art in order to combine prior art
references, examiners were apparently quite reluctant to combine life
science and software prior art. This reluctance appeared even though
the applications in question, which had an average of 3.36 inven-
tors,162 presumably often included both inventors with biological and
software skills.
157. This is assuming that the matched set of applications coming into the two art units
had roughly similar attributes with respect to notice (an assumption that seems likely except
that applications in AU 1631 may have been likely to have more problems with respect to
multiple inventions in the same patent).
158. See supra Table 3.
159. 66 Fed. Reg. at 1092.
160. See supra Table 3.
161. Id.
162. See supra Table 2.
No. 1] Bioinformatics at the Patent Office 237
In Part V, we discuss further the policy implications of our find-
ings for patent quality. We also discuss the normative question of how,
in view of these empirical findings, patent applications on interdisci-
plinary, team-based inventions should be examined. We propose that
the answer turns in part on when the interdisciplinarity of the field in
question itself becomes relatively routine. In the case of bioinformat-
ics, this may have occurred before 2003. However, given the Federal
Circuit’s pre-KSR demand of a written teaching, suggestion, or moti-
vation requirement, examiners may have been reluctant to combine
life science and software references.
V. POLICY IMPLICATIONS FOR PATENT QUALITY AND
TEAM-BASED INNOVATION
Given our relatively specific empirical focus on two art units, we
tread carefully when it comes to implications for the patent system as
a whole. However, our results do have implications for two areas:
patent quality and examination of team-based innovation.
A. Patent Quality
As for quality, our results suggest that some combination of ex-
aminer training and advanced educational background may have an
impact on patent quality, particularly with respect to notice. Thus,
efforts to train examiners, particularly in the use of the written de-
scription requirement, and in mechanisms by which patent applicants
can be forced to specify the meaning of potentially problematic claim
terms,163 are likely to prove fruitful. Moreover, while the USPTO is
unlikely to attract significant numbers of individuals with advanced
degrees in most art units, training examiners in basic principles of
scientific peer review may prove useful as well.
B. Team-Based Innovation
Our inquiry also has implications for patent examination in an era
of team-based, interdisciplinary science. Scientific knowledge produc-
tion is increasingly team-based; indeed team sizes have risen at an
average rate of 15% to 20% per decade, and this increase appears in
nearly all subfields of research and invention.164 Patent applications
163. See Menell & Meurer, supra note 58 at 37–39.
164. Benjamin Jones, As Science Evolves, How Can Science Policy?, in 11 INNOVATION
POLICY AND THE ECONOMY 103, 113 (Josh Lerner & Scott Stern eds., 2011).
238 Harvard Journal of Law & Technology [Vol. 29
have mirrored this shift. All areas of patenting have increased in team
size over the past 25 years across all countries.165
The recognition that knowledge production is increasingly team-
based has prompted some scholars to call for a doctrinal shift away
from the familiar legal reference point of a “person having ordinary
skill in the art” (“POSITA”) to a “team having ordinary skill in the art”
(“TOSITA”). In a 2002 article, for example, Joseph Meara suggested
that in fields where advances are typically made in interdisciplinary
teams, a team-based standard would be more appropriate than an indi-
vidual standard.166 Meara gave the example of implementing a Dutch
auction on the Internet. In that example, a team consisting of a soft-
ware engineer and a businessperson with MBA training would pre-
sumably have found the idea obvious, even though either individual
alone might not have.
Although Meara’s proposal was promulgated prior to KSR v. Te-
leflex, the principles of that proposal apply with even greater force
after KSR. In a 2011 article advocating an inducement standard for
nonobviousness (under which patents would be granted only on those
inventions that “would not be disclosed or devised but for the in-
ducement of a patent”),167 Michael Abramowicz and John Duffy en-
dorsed making the inducement determination at the level of the
inventive team.168
We build upon this earlier work, but propose a slightly different
approach. In our view, a team-based approach should be used when a
field has become routinely interdisciplinary. This is because the very
act of creating a team may be innovative when it brings two previous-
ly disparate fields together into a new combinatorial space. Moreover,
although a patent grant may not always be a sign of true invention (for
the reasons we have discussed at length), at least some of these pa-
tents were presumably granted at a time when the interdisciplinarity in
question was still nascent.
Recent empirical work has begun to quantify this combinatorial
process of invention. Notably, a new study of U.S. patent and tech-
nology classification records from 1790 to 2010 demonstrates that
patenting over that time has been characterized not only (or even pri-
marily) by the creation of new technological capabilities but by the
increasingly complex combination of existing technological building
165. Id.; see also Zhenzhong Ma & Yender Lee, Patent Application and Technological
Collaboration in Inventive Activities: 1980–2005, 28 TECHNOVATION 379, 388 (2008).
166. See Joseph P. Meara, Just Who Is the Person Having Ordinary Skill in the Art? Pa-
tent Law’s Mysterious Personage, 77 WASH. L. REV. 267, 293 (2002).
167. Michael Abramowicz and John F. Duffy, The Inducement Standard of Patentability,
120 YALE L.J. 1590, 1590 (2011) (quoting Graham v. John Deere Co., 383 U.S. 1, 11
(1966)).
168. Id. at 1615.
No. 1] Bioinformatics at the Patent Office 239
blocks.169 Because the USPTO assigns relevant USPC classifications
to each patent,170 a patent’s classes identify the distinct technologies
that the inventor combined to produce the invention — and the com-
bination identifies the particular interdisciplinarity at work in that in-
stance of inventive activity.171 Historically, the rate at which new
inventions have introduced new technological capabilities, represent-
ing new technological classes, has slowed considerably.172 Yet sur-
prisingly, the rate at which new combinations of technological classes
have emerged has systematically kept pace with the number of new
patents.173
These results suggest that whereas some inventions represent new
combinations of technological capabilities, other inventions represent
merely existing combinations of technological capabilities. The for-
mer shows emerging interdisciplinarity; the latter, routine investiga-
tion within an increasingly well-defined field. A prior empirical study
demonstrated that there is a 60% likelihood that a given invention
augurs a new technological combination and only a 40% likelihood
that it relies on an existing technological combination.174 Thus, inter-
disciplinarity is, and historically has been, the prevailing mode of in-
novation.
To be sure, our proposed doctrinal inquiry of routine interdisci-
plinarity may sometimes prove difficult to implement. Beyond the
ordinary line-drawing problems that inform all such taxonomic de-
terminations, there is the added temporal difficulty of determining
when previously unrelated disciplines should be regarded as solidly
linked.
In the particular case of bioinformatics, we proceed with the ben-
efit of a historical record. The systematic collection and analysis of
biological sequence data has commanded the collaborative efforts of
“computer scientists, statisticians, and biologists” for over four dec-
ades,175 and the term “bioinformatics” itself dates from 1970.176 The
idea of a bioinformatician proper, however, is of more recent vintage,
dating between the mid-1990s177 and the early 2000s.178 During this
169. Hyejin Youn et al., Invention as a Combinatorial Process: Evidence from US Pa-
tents, 12 J. ROYAL SOC’Y INTERFACE 20150272, at *3 (2015),
http://rsif.royalsocietypublishing.org/content/12/106/20150272.full-text.pdf
[http://perma.cc/2P4M-7ERF].
170. See supra note 103 and accompanying text.
171. Youn et al., supra note 169, at *3–4.
172. Id. at *4.
173. Id.
174. Id. at *5.
175. Andrew Chin, Research in the Shadow of DNA Patents, 87 J. PAT. & TRADEMARK
OFF. SOC’Y 846, 894 (2005).
176. Paulien Hogeweg, The Roots of Bioinformatics in Theoretical Biology, 7 PLOS
COMPUT. BIOL. e1002021 at *1 (2011).
177. See Contreras, supra note 6 and accompanying text.
240 Harvard Journal of Law & Technology [Vol. 29
time, public accounts began referring to bioinformatics as its own
field.179 Moreover, universities, such as the University of Michigan180
and the University of California San Diego,181 began establishing
training and degree programs in bioinformatics.182
To the extent of its authority, the USPTO worked to stay abreast
of this shift toward team-based scientific research. The agency estab-
lished its art unit in bioinformatics precisely as the fields of biology
and software were becoming solidly linked. Indeed, as previously not-
ed, by the year 2003 (the year that we studied), the role of computer
science in biology was sufficiently well-established that Lincoln Stein
predicted bioinformatics had only “ten years to live” as a field that
could be considered interdisciplinary in the first instance.183 However,
the shadow of the Federal Circuit’s very exacting requirements for
combining prior art may have kept AU 1631 from fully assimilating
team-based research norms and practices. This intuition could be test-
ed, at least to some extent, by analyzing the frequency of obviousness-
based rejections in AU 1631 after the Supreme Court decision in KSR
v. Teleflex.
VI. CONCLUSION
The empirical data on bioinformatics examination that this Article
has presented may offer important contributions to the literature on
patent quality. We have shown that patent applications in bioinformat-
ics generally received more stringent examiner scrutiny and more re-
jections, particularly on notice-related grounds, than did applications
in conventional software informatics. Bioinformatics examiners paid
significant attention to notice not only in the biological aspects of the
inventions that they examined, but also in the inventions’ strictly in-
178. See Charles Vorndran & Robert L. Florence, Bioinformatics: Patenting the Bridge
Between Information Technology and the Life Sciences, 42 IDEA 93, 126 (2002) (noting
that “bioinformatics is a multidisciplinary field” and that “the field, itself, is fairly new, yet
developing rapidly”).
179. E.g., Brad Stone, Wanted: Hot Industry Seeks Supergeeks, NEWSWEEK (Apr. 29,
2001), http://www.newsweek.com/wanted-hot-industry-seeks-supergeeks-150551
[https://perma.cc/642P-ZGM7].
180. UNIVERSITY OF MICHIGAN, U-M Bioinformatics Graduate Program, DEPARTMENT
OF COMPUTATIONAL MEDICINE & BIOINFORMATICS,
http://www.ccmb.med.umich.edu/graduate-program [http://perma.cc/VVA5-63XR] (dating
the first class to 2001).
181. UNIVERSITY OF CALIFORNIA AT SAN DIEGO, Bioinformatics and Systems Biology
Graduate Program, UC SAN DIEGO, http://www.ucsd.edu/catalog/curric/BIOI-gr.html
[http://perma.cc/E8ZN-35B8] (dating the program’s founding to 2001).
182. Stone, supra note 179 (noting that, as of 2001, schools including the University of
California Davis, the University of California Berkeley, and Cornell University had estab-
lished bioinformatics programs or were planning to do so).
183. Steinberg, supra note 127.
No. 1] Bioinformatics at the Patent Office 241
formational aspects. Our results suggest that institutional investment
in human capital and educational training can make a difference.
Our results did not hold, however, for application of the nonobvi-
ousness requirement before the KSR decision. KSR may therefore
have been a particularly important precedent for proper evaluation of
interdisciplinary and team-based science.
Our Article also offers the first empirical account of patent qual-
ity at the juxtaposition of two fields generally believed to be wholly
opposed with respect to patent quality, particularly boundary notice.
At the stage of initial application, we found significant evidence that
bioinformatics applications were “better” than conventional informat-
ics applications.
Finally, our results invite further study of invention that is con-
ducted in collaborative environments and draws on expertise in a va-
riety of disciplines, as bioinformatics does. A richer empirical account
of these inventive activities would do much to align the U.S. patent
system with the modern realities of team-based innovation.
