Fig. 5: Ras/Ca
Signaling in DALD.
(only AGP I shown)
Darker colors indicate higher levels of significance. Pink
circle: growth factor regulation, green circle: ion chan-
nels, red boxes: familal hemiplegic migraine (FHM) genes.
Fig. 2: Genetic Risk Model. Neighboring SNPs can be
dependent and are in LD with intermediate disease loci,
unless separated by a recombination hotspot (×).
Fig. 3: Study-Specific GW Significance. The cut-off
(solid bar) is estimated as the median projection from the
chromosomes with the lowest deviation from the
projection (dashed curves/bars). (partial results shown)
Calculating ‘study-specific’ genome-wide signifi-
cance accounts for (i) the most significant results
to require high MAF and (ii) for
tests being performed within a sliding window.
Finding the Missing Heritability in GWAS of Epidemiological Studies and Clinical Trials:
Repurposed Drugs for Autism, Crohn’s Disease, and Breast Cancer
The Rockefeller University Hospital, Center for Clinical and Translational Science, Biostatistics, Epidemiology & Research Design (BERD)
1. Access to the muGWAS clou d/GPU infrastructure is available as a
Web-service at http://mustat.rockefeller.edu.
2. Wittkowski KM, et al. (2013) Pharmacogenomics 14(4): 391.
Available from: http://www.ncbi.nlm.nih.gov/pubme d/23438886
3. Wittkowski KM, et al. (2014) Translational Psychiatry 4:e354.
Available from: http://www.nature.com/articles/tp2013124 .
Fig. 6: Childhood Absence Epilepsy (CAE) in 185
children vs. 370 Illumina controls.
In a phase 3 trial, only 400 of 600 treated sub-
jects responded to the study drug with respect
to either of two outcomes. While ssGWAS
results were inconclusive Fig. 12, only outcome I shown,
muGWAS implicated mutations along a pathway
not targeted by the drug. Fig. 13
A: Outcome 1
Fig. 4: High enrichment of µGWAS vs ssGWAS.
Some non-responders have a mutation in the drug target
(left), but most have mutations in genes along a different
pathway (right) and, thus, need a different drug.
Supported in par t by grant # UL1 TR000043 from the National Center for Advancing
Translational Scien ces (NCATS, National Institutes of Health (NIH) Clinical Transla-
tional Science Award (CTSA) program), by grant # 24481 32 from the Simons Founda-
tion Autism Research In itiative, and by a grant from the Center for Basic and Transla-
tional Research on Disorders of the Digestive System. KMW is inventor/assignee of re -
lated patent(application)s and a consultant to Johnson&Johnson on one of the projects.
Methods I (muGWAS)
Methods II (Decision Strategy)
Introduction AUTISM (EPILEPSY) CROHN’S DISEASE BREAST CANCER Conclusion (Utilizing dbGaP)
SNP.A SNP.X SNP.Y SNP.Z
recessive: aa = aA < AA
dominant: aa < a A = AA
allelic: a a=0, aA=1, AA=2
ordinal: aa < aA < AA
In two studies of 1071/576 autism cases (AGP
1/II,Anney 2010/12) the significant genes for compar-
ing non-verbal to verbal subjects Fig. 5were
highly consistent with previous CAE results,Fig. 6
and included genes known to cause familial
hemiplegic migraines (FHM).Fig. 4
The mutism-specific role of K+/Cl−channels
suggests MFAFig. 4 (also effective against mi-
graines) as an orphan drug to prevent DALD
(Disruption of Active Language Development).3
A decade after the Human Genome Project, a
novel computational biostatistics approach finds
the ‘missing heritability’ in GWAS, suggesting
reformulated drugs to prevent mutism in
autism, recurrence in Crohn's Disease, and
metastases in breast cancer.
With replicated results from 600–2300 subjects
only, one can now also detect risk factors for
treatment failure in clinical trials and genetic
susceptibility in emerging diseases, such as Zika.
Adopting statistical methods and decision stra-
tegies to genetics (GWAS)3yields novel treat-
ments for common diseases from 10 years of
data available, e.g., in the NIH’s dbGaP.
Fig. 12: µGWAS results
implicating a gene asso-
ciated with outcome I.
A: ssGWAS results for out-
come I are inconclusive.
B: µGWAS results for out-
come I implicate a gene at
around 100 Mb. Black dots
are ssGWAS, diamonds are
muGWAS results by diplo-
type width (size) and infor-
mation content (red: low).
C: Regional Manhattan Plot.
Linetype indicates DT width
(dotted: 2 .. solid 6 SNPs).
D: No association in this
LD block with outcome II.
E: HapMap LD (red) and re-
comb. rate (blue). The peak
is located in the regultory
region of a known risk gene.
B: Outcome I
C: Outcome I
D: Outcome II
13: High enrichment of muGWAS/muPhene
vs ssGWAS/Outcome I. Some non-responders have a
Mutation in the drug target (left), but most have muta-
tions along a diffe-
Statistical tests emerged only with the advent of
and were initially (20
century) limited to the gen. linear model (GLM).
Fig. 1: Statistics and
Memory. Mechanic: t-
stepwise regression /
PCA, PC: MC / Bayes.
In GWAS, the GLM
often create artifacts
(‘faces in the sky’).
U-statistics for multivariate data became feasible
only in 2001 (32bit OS: GB). In contrast to single-
SNP GWAS (ssGWAS) and linear weight (‘allelic’)
or lin/log regression GWAS, muGWAS avoids
artifacts and increases power by accounting for
genetics (LD structure, varying dominance, ...).
Fig. 9: QQ-Plots
and Genes by Sig-
; blue line:
genes significant in
included in Fig. 8).
Fig. 8: Impaired Fucosylation as a Common Risk
Factor in Crohn’s Disease.
Fig. 4: Mutism in Autism as Channelopathy.
Mutism mutations increase neuronal K
facilitates adjustment (hyperpolarization) to stress
(‘stranger anxiety’ by activating outward K
In 1000 subjects,(Duerr 2006) muGWAS confirmed
many of the results from later studies of
32/70,000 (Jostin 2012/Liu 2015) subjects. Except for
two genes essential for fucosylation, GMDS and
ENTPD5,Fig. 7/9, yellow all significant genes could be
related to known CD pathways.Fig. 7/9, blue
Together with FUT2 as a known risk factor, the
results suggest that 50% of CD patients might
benefit from L-fucose supplementation, a safe
intervention, whicfh is effective in LAD II.Fig. 7
7: The Fucose Salvage Pathway. Supplementation
with nutritional L-Fucose can compensate for defects in
the L-fucose-GDP/GMP antiporter SLC35C1.
11: Breast Cancer Risk Genes. All three studies
implicate functionally similar membrane-signaling (yellow)
and cell-cycle control (blue) genes (see Fig. 10 for ).
Fig. 8: Hypothetical Interaction Between Common
Risk Factors in Crohn’s Disease. Genes highlighted
are significant in muGWAS (ATG16L1: ssGWAS only).