Automatically extracting functionally equivalent proteins from SwissProt.

Abstract

There is a frequent need to obtain sets of functionally equivalent homologous proteins (FEPs) from different species. While it is usually the case that orthology implies functional equivalence, this is not always true; therefore datasets of orthologous proteins are not appropriate. The information relevant to extracting FEPs is contained in databanks such as UniProtKB/Swiss-Prot and a manual analysis of these data allow FEPs to be extracted on a one-off basis. However there has been no resource allowing the easy, automatic extraction of groups of FEPs - for example, all instances of protein C.We have developed FOSTA, an automatically generated database of FEPs annotated as having the same function in UniProtKB/Swiss-Prot which can be used for large-scale analysis. The method builds a candidate list of homologues and filters out functionally diverged proteins on the basis of functional annotations using a simple text mining approach.
Large scale evaluation of our FEP extraction method is difficult as there is no gold-standard dataset against which the method can be benchmarked. However, a manual analysis of five protein families confirmed a high level of performance. A more extensive comparison with two manually verified functional equivalence datasets also demonstrated very good performance.
In summary, FOSTA provides an automated analysis of annotations in UniProtKB/Swiss-Prot to enable groups of proteins already annotated as functionally equivalent, to be extracted. Our results demonstrate that the vast majority of UniProtKB/Swiss-Prot functional annotations are of high quality, and that FOSTA can interpret annotations successfully. Where FOSTA is not successful, we are able to highlight inconsistencies in UniProtKB/Swiss-Prot annotation. Most of these would have presented equal difficulties for manual interpretation of annotations. We discuss limitations and possible future extensions to FOSTA, and recommend changes to the UniProtKB/Swiss-Prot format, which would facilitate text-mining of UniProtKB/Swiss-Prot.

Full text

Page 1

BioMed Central
Page 1 of 14
(page number not for citation purposes)
BMC Bioinformatics
Open Access
Database
Automatically extracting functionally equivalent proteins from
SwissProt
Lisa EM McMillan and Andrew CR Martin*
Address: Research Department of Structural & Molecular Biology, University College London, Gower Street, London WC1E 6BT, UK
Email: Lisa EM McMillan - mcmillan@biochem.ucl.ac.uk; Andrew CR Martin* - andrew@bioinf.org.uk
* Corresponding author
Abstract
Background: There is a frequent need to obtain sets of functionally equivalent homologous
proteins (FEPs) from different species. While it is usually the case that orthology implies functional
equivalence, this is not always true; therefore datasets of orthologous proteins are not appropriate.
The information relevant to extracting FEPs is contained in databanks such as UniProtKB/Swiss-
Prot and a manual analysis of these data allow FEPs to be extracted on a one-off basis. However
there has been no resource allowing the easy, automatic extraction of groups of FEPs – for
example, all instances of protein C.
We have developed FOSTA, an automatically generated database of FEPs annotated as having the
same function in UniProtKB/Swiss-Prot which can be used for large-scale analysis. The method
builds a candidate list of homologues and filters out functionally diverged proteins on the basis of
functional annotations using a simple text mining approach.
Results: Large scale evaluation of our FEP extraction method is difficult as there is no gold-
standard dataset against which the method can be benchmarked. However, a manual analysis of five
protein families confirmed a high level of performance. A more extensive comparison with two
manually verified functional equivalence datasets also demonstrated very good performance.
Conclusion: In summary, FOSTA provides an automated analysis of annotations in UniProtKB/
Swiss-Prot to enable groups of proteins already annotated as functionally equivalent, to be
extracted. Our results demonstrate that the vast majority of UniProtKB/Swiss-Prot functional
annotations are of high quality, and that FOSTA can interpret annotations successfully. Where
FOSTA is not successful, we are able to highlight inconsistencies in UniProtKB/Swiss-Prot
annotation. Most of these would have presented equal difficulties for manual interpretation of
annotations. We discuss limitations and possible future extensions to FOSTA, and recommend
changes to the UniProtKB/Swiss-Prot format, which would facilitate text-mining of UniProtKB/
Swiss-Prot.
Background
It is often necessary to compare the 'same' gene or gene
product (protein) in different species. By the 'same' pro-
tein, we mean an orthologue that performs an equivalent
function or functions. Obtaining lists of functionally-
equivalent proteins (FEPs) is fundamental for compara-
Published: 6 October 2008
BMC Bioinformatics 2008, 9:418doi:10.1186/1471-2105-9-418
Received: 3 March 2008
Accepted: 6 October 2008
This article is available from: http://www.biomedcentral.com/1471-2105/9/418
© 2008 McMillan and Martin; licensee BioMed Central Ltd.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0),
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Page 2

BMC Bioinformatics 2008, 9:418http://www.biomedcentral.com/1471-2105/9/418
Page 2 of 14
(page number not for citation purposes)
tive and evolutionary genomics, and downstream pro-
teomic studies [1]. The particular motivation for the
current work was obtaining lists of FEPs to examine resi-
due conservation scores and to aid in understanding the
effects of mutations on protein function in the context of
a large-scale automated analysis pipeline, SAAPdb [2].
Proteins that have diverged in function (either by gaining
or losing functionality) will show differences at key func-
tional residues. We therefore needed a reliable automatic
method for extracting groups of FEPs from UniProtKB/
Swiss-Prot.
Consider, for example, the HOX family of genes, which is
a large family of transcription factor proteins containing
the well characterised homeobox motif. These proteins
are well conserved across species and are believed to be
critical in embryogenesis, oncogenesis and differentiation
processes such as haematopoiesis [3,4]. HOX proteins are
representative of large protein families in that there are
several paralogues within a species (thirteen in the case of
the human HOX family [3]), and each paralogue can be
involved in several distinct aspects of the same biological
process. A sequence alignment of such evolutionarily
related, but functionally different, proteins would contain
significant noise, and obscure much of the genuine func-
tional conservation between true FEPs.
While homology does not imply functional equivalence,
it is also not possible to use functional data alone to iden-
tify FEPs. Proteins can converge on similar functions with-
out being evolutionarily related. For example, subtilisin
(EC 3.4.21.62) and trypsin (EC 3.4.21.4) have evolved
separately in bacteria and vertebrates respectively; they
differ significantly in protein sequence, structure and fold,
yet the same three amino acids form the catalytic triad in
both proteins [5]. Aligning such functionally similar, but
evolutionarily unrelated, proteins is meaningless; we are
interested in proteins which are both homologous and
functionally equivalent.
Two entities are homologous if they have a common evo-
lutionary origin. An orthologous relationship denotes that
this common origin was a speciation event, whereas para-
logues are related by a gene duplication [6]. Paralogues,
having been derived via a mechanism for functional diver-
gence, are likely to perform different functions [7]. While
orthologues generally perform the same function, it is
possible for the function to diverge, particularly when
orthologues are evolutionarily distant [6]. For example,
Shibata et al. [8] showed that although the general func-
tion of exportin-5 proteins (nuclear export of miRNAs and
tRNAs) is conserved across different species, substrate spe-
cificity varies. Further, the AGAMOUS gene in Arabidopsis
is involved in carpel and stamen development, but the
two orthologues in maize have specialised: ZAG1 is highly
expressed during carpel development, and ZMM2 is
expressed during stamen development [9]. It is clear then
that orthology need not imply strict functional equiva-
lence, and it follows that sets of orthologues, defined by
methods such as Inparanoid [10], C/KOG [11,12] and
TOGA [13], are not appropriate as lists of FEPs. Further,
these methods are computationally intensive and as such
are often limited to small species sets.
The identification of true FEPs requires consideration of
features such as functional assays, interaction networks,
expression data and so forth. UniProtKB/Swiss-Prot is a
carefully annotated databank of protein sequences that
includes functional annotations. While many of these are
transferred through orthology, where there is experimen-
tal evidence for function, it will be included. Thus, short
of conclusive experimental studies, the most reliable way
of identifying families of FEPs is first to identify families
of homologues in UniProtKB/Swiss-Prot and then to
examine the annotations to find a set of proteins that are
annotated as performing the same function or functions.
It is, of course, possible that annotations in UniProtKB/
Swiss-Prot will be incorrect, but as UniProtKB/Swiss-Prot
is updated on a regular basis, it is expected that these
annotations will represent the most up-to-date state of our
knowledge of protein function, and errors in annotations
will be corrected with future releases.
While it is perfectly possible to perform this analysis on an
individual basis by searching UniProtKB/Swiss-Prot for
homologues and comparing the annotations manually,
there is a pressing need for an automatically updated
resource that simply lists families of FEPs in UniProtKB/
Swiss-Prot. Several methods exist that exploit database
annotations to identify related proteins [14-17], however
there has been no resource that very simply provides sets
of FEPs annotated as having the same function in Uni-
ProtKB/Swiss-Prot in an easily-accessible format, with
extensive coverage of multiple proteomes.
We have developed FOSTA (Functional Orthologues from
SwissProt Text Analysis), which automates the process
that one would perform manually to extract a family of
FEPs from UniProtKB/Swiss-Prot. It considers UniProtKB/
Swiss-Prot proteins for inclusion in groups of FEPs
(FOSTA families) rooted around human proteins. It
refines an initial candidate list of homologues on the basis
of functional annotation similarity, to distinguish FEPs
from functionally diverged homologues (FDHs). To
assess functional annotation similarity, we employ simple
text-mining techniques to compare UniProtKB/Swiss-Prot
description fields.

Page 3

BMC Bioinformatics 2008, 9:418http://www.biomedcentral.com/1471-2105/9/418
Page 3 of 14
(page number not for citation purposes)
Results and discussion
Evaluating FOSTA is difficult because no gold-standard
dataset exists. In addition, it is difficult to design an eval-
uation procedure to isolate the performance of FOSTA
itself from the quality of the UniProtKB/Swiss-Prot anno-
tations that FOSTA interprets. To assess the FOSTA method,
we need to assess whether FOSTA is grouping proteins
correctly into functionally equivalent groups given the
functional annotations, rather than assessing whether the
functional annotations are of sufficient detail to infer gen-
uine functional equivalence. However, it is also very
important to assess the latter, as FOSTA is dependent on
the UniProtKB/Swiss-Prot annotations.
As such, FOSTA has been evaluated in three phases. The
first involves manual interpretation of the results of sev-
eral large protein families, some chosen at random, and
some chosen as known problematic cases. This phase
assesses how well FOSTA can interpret functional annota-
tions, and infer functional equivalence compared with
manual interpretation. The second phase benchmarks
FOSTA against a fully manually annotated dataset, and a
larger partially annotated dataset. This phase not only
indicates whether FOSTA performs well, but also assesses
whether the annotations are good enough to infer func-
tional equivalence. The final phase of evaluation involves
comparing UniProtKB/Swiss-Prot with InParanoid [10], a
popular method for identifying orthologues.
FOSTA results are available at http://www.bioinf.org.uk/
fosta/, by searching with the UniProtKB/SwissProt protein
ID of interest. Results in this paper are for UniProtKB/
Swiss-Prot version 53.0 (29th May 2007). The full set of
FOSTA results may also be obtained in XML format, as can
the results for a single human protein. A comprehensive
help service is provided online, and updates will be per-
formed every two months.
An overview of FOSTA families
Before presenting the analysis of our method, we provide
a brief description of the dataset. With a view to summa-
rising the FOSTA dataset, we have calculated the 'Uni-
ProtKB/Swiss-Prot proteome coverage' for each species
described in FOSTA. This has been calculated as NF/NSP,
where NSP is the number of proteins from that species that
are described in UniProtKB/Swiss-Prot (i.e., the size of the
'UniProtKB/Swiss-Prot proteome') and NF is the number
of proteins from that species described in FOSTA. There-
fore, a species which is fully represented in FOSTA with
respect to its UniProtKB/Swiss-Prot proteome would have
a UniProtKB/Swiss-Prot proteome coverage of 100%,
while a species with none of its UniProtKB/Swiss-Prot
proteins represented in FOSTA would have a UniProtKB/
Swiss-Prot proteome coverage of 0%. Of the 11126 spe-
cies represented in UniProtKB/Swiss-Prot version 53.0,
just over half (52.73%) are not represented by FOSTA.
This will in part be due to differing annotation formats of
very remote species, but will also in part be due to distant
species having very few proteins in common with the
Human proteome. More positively, 2550 species
(22.92%) are fully represented in FOSTA. Of course, many
of these proteomes will be small, but nevertheless, it is
encouraging that almost a quarter of UniProtKB/Swiss-
Prot species are fully represented in FOSTA.
The most common family size is two: 25.48% (3793/
14884) of FOSTA families with a non-human member
have two members; this usually corresponds to an exclu-
sively Human/Murine FOSTA family. These are not only
the most well represented species in UniProtKB/Swiss-
Prot version 53.0, they are also the most extensively and
similarly annotated. 37.25% of FOSTA families (5545)
have five or more members, and only 1.85% (275) have
more than 50 members.
With respect to how FEP relationships are formed, most
FOSTA families are formed exclusively using the protein
prefix match, i.e., all members share the same protein pre-
fix. However, 42.10% (6266/14884) of FOSTA families
contain at least two different protein prefixes. Further-
more, of the 22 871 protein prefixes recorded in FOSTA,
5.42% are found to exist in more than one FOSTA family.
This indicates that, although UniProtKB/Swiss-Prot pro-
tein prefixes are very often reliable, incorporating addi-
tional information derived from the description field is
beneficial in identifying FEP relationships.
HOX proteins
In the introduction, we discussed the family of HOX pro-
teins as an example of a large family of proteins with mul-
tiple paralogues in each species. Here we assess the
performance of FOSTA (and – by proxy – the quality of
UniProtKB/Swiss-Prot annotations) when assigning the
Zebrafish (Danio rerio) FEP to Homo Sapiens homeobox
protein Hox-B7. There is a body of literature on the prob-
lem of elucidating HOX gene evolution, which is difficult
in Zebrafish given the extensive polyploidy in its evolu-
tionary history [18-20].
The BLAST search identifies 83 Zebrafish candidate FEPs
and the filtering process assigns HXB7A_DANRE [Swiss-
Prot:Q8AWY9] to the FOSTA family of HXB7_HUMAN
[Swiss-Prot:P09629]. There are 24 Zebrafish FDHs that
have higher sequence similarity to HXB7_HUMAN than
the assigned FEP. These proteins, the FEP and the root
human protein are listed in Table 1, along with their Uni-
ProtKB/Swiss-Prot annotations and their sequence iden-
tity to HXB7_HUMAN. It is clear that HXB7A_DANRE
should be identified as the FEP given the similarity of its
description to that of HXB7_HUMAN; this would be

Page 4

BMC Bioinformatics 2008, 9:418http://www.biomedcentral.com/1471-2105/9/418
Page 4 of 14
(page number not for citation purposes)
selected in a manual analysis of these candidates, despite
its lower sequence identity.
Several sites of functional relevance have been identified
for HXB7_HUMAN (Table 2). These functional sites have
been extracted from UniProtKB/Swiss-Prot annotations
and a mutagenesis study by Yaron et al. (2001). Figure 1
shows the alignment of the HXB7_HUMAN and the four
confidently assigned FEPs with HXB7A_DANRE and the
other fifteen Danio rerio candidates in the functionally rel-
evant areas. Despite globally having the lowest sequence
identity to HBX7_HUMAN of all the Zebrafish proteins
shown in Figure 1, it is clear that HXB7A_DANRE has the
highest conservation at functionally critical sites. Across
residues 126 to 133, HXB7A_DANRE only differs from
HXB7_HUMAN at the position of a putative PBX binding
site, unlike the other Danio rerio proteins which all differ
in a known sequence motif. The homeobox region (which
also includes crosslinking sites) is highly conserved across
all of the Zebrafish proteins, and again, conservation is
highest in HXB7A_DANRE. None of the Zebrafish pro-
teins show conservation at residues 203 and 204, which
describe a putative CKII target site [3]. It is possible that
this functional site has been wrongly predicted; however,
this is unlikely as it is absolutely conserved across the five
mammalian species. It is more likely that this region is no
longer functional in the Danio rerio lineage, or that this is
a recently acquired functionality in the mammalian clade.
A solved annotation problem: PROC_HUMAN
The UniProtKB/Swiss-Prot ID consists of a protein name
followed by an underscore and the species name. It was
our initial assumption that the protein name part of the
ID was a unique name used to label FEPs [21]. However,
while analysing human protein C (PROC_HUMAN)
using UniProtKB/Swiss-Prot version 50.6, it was evident
that this approach was unreliable. The 'PROC' prefix was
in forty different species to describe three different pro-
teins: Procalin in one species (PROC_TRIPT, [Swiss-
Prot:Q9U6R6]), protein C in 11 species (e.g.,
PROC_HUMAN, [Swiss-Prot:P04070]), and pyrroline-5-
carboxylate reductase in the remaining 28 species (e.g.,
PROC_ECOLI, [Swiss-Prot:P0A9L8]) (UniProtKB/Swiss-
Prot version 53.0). A previous version of FOSTA was suc-
cessful in correctly assigning only true examples of protein
C to the FEP group, and analysis of human pyrroline-5-
corboxylate reductase results highlighted the inconsisten-
Table 1: Zebrafish candidates for the FOSTA family of HXB7_HUMAN
ProteinID Description
HXB7_HUMAN
HXB7A_DANRE
100
54
Homeobox protein Hox-B7; Hox-2C; HHO.C1
Homeobox protein Hox-B7a; Hox-B7
HXA1A_DANRE
HXA3A_DANRE
HXA4A_DANRE
HXA5A_DANRE
HXA9B_DANRE
HXB1A_DANRE
HXB1B_DANRE
HXB2A_DANRE
HXB3A_DANRE
HXB4A_DANRE
HXB5A_DANRE
HXB5B_DANRE
HXB6A_DANRE
HXB6B_DANRE
HXB8B_DANRE
HXC1A_DANRE
HXC3A_DANRE
HXC5A_DANRE
HXC6A_DANRE
HXC6B_DANRE
HXC8A_DANRE
HXD4A_DANRE
HXD9A_DANRE
HXDAA_DANRE
63
68
65
75
62
64
64
57
67
62
75
75
78
75
60
62
61
72
63
77
73
62
65
61
Homeobox protein Hox-A1a; Hox-A1
Homeobox protein Hox-A3a
Homeobox protein Hox-A4a; Zf-26; Hoxx4
Homeobox protein Hox-A5a
Homeobox protein Hox-A9b
Homeobox protein Hox-B1a; Hox-B1
Homeobox protein Hox-B1b; Hox-A1
Homeobox protein Hox-B2a; Hox-B2
Homeobox protein Hox-B3a; Hox-B3
Homeobox protein Hox-B4a; Hox-B4; Zf-13
Homeobox protein Hox-B5a; Hox-B5; Zf-21
Homeobox protein Hox-B5b; Hox-B5-like; Zf-54
Homeobox protein Hox-B6a; Hox-B6; Zf-22
Homeobox protein Hox-B6b; Hox-A7
Homeobox protein Hox-B8b; Hox-A8
Homeobox protein Hox-C1a
Homeobox protein Hox-C3a; Hox-114; Zf-114
Homeobox protein Hox-C5a; Hox-C5; Hox-3.4; Zf-25
Homeobox protein Hox-C6a; Hox-C6; Zf-61
Homeobox protein Hox-C6b
Homeobox protein Hox-C8a
Homeobox protein Hox-D4a; Hox-D4
Homeobox protein Hox-D9a; Hox-D9
Homeobox protein Hox-D10a; Hox-D10; Hox-C10
Protein: The UniProtKB/Swiss-Prot ID; ID: The sequence identity of the Protein to HXB7_HUMAN;
Description: The UniProtKB/Swiss-Prot description (DE) field.

Page 5

BMC Bioinformatics 2008, 9:418 http://www.biomedcentral.com/1471-2105/9/418
Page 5 of 14
(page number not for citation purposes)
cies in UniProtKB/Swiss-Prot naming conventions (data
not shown).
Several of the FEPs in the FOSTA families of
P5CR1_HUMAN (pyrroline-5-carboxylate reductase 1)
and PROC_HUMAN (protein C) have had multiple pro-
tein prefix changes. However, after notifying UniProtKB/
Swiss-Prot of the discrepancies, all the misnamed proteins
were corrected for the release of UniProtKB/Swiss-Prot
v51.2: pyrroline-5-carboxylate reductase proteins prefixed
with PROC or PROH are now prefixed with P5CR or
P5CR1 and PROC_TRIPT (procalin) is now called
PRCLN_TRIPT.
Verifying the Danio rerio FEP of HXB7_HUMAN using annotated functional regions
Figure 1
Verifying the Danio rerio FEP of HXB7_HUMAN using annotated functional regions. Residues identical to that of
HXB7_HUMAN are in bold capitals and highlighted yellow, mismatching residues are non-captials and highlighted in light grey.
The root human protein (HXB7_HUMAN) is indicated in the red box, and the assigned Zebrafish is highlighted in the blue box.
The position relative to HXB7_HUMAN is given on the top line, and the asterisks on the bottom line highlight fully conserved
columns.
Table 2: Functional sites in HXB7_HUMAN
Functional siteLocation Reference
DNA binding (homeobox)
Crosslink (glycyl lysine isopeptide)
Motif (Antp-type hexapeptide)
Hypothesized binding to PBX
Putative CKII target
Putative CKII target
137 – 197
191 & 193
126 – 131
129 – 130
132 – 133
203 – 204
UniProtKB/Swiss-Prot FT/DNA_BIND annotation
UniProtKB/Swiss-Prot FT/CROSSLNK annotation
UniProtKB/Swiss-Prot FT/MOTIF annotation
Yaron et al. [3]
Yaron et al. [3]
Yaron et al. [3]
Functional site: a description of the functional site; Location: the residue number in HXB7_HUMAN;
Reference: The source of the annotation.