Comparative network analysis of neurological disorders focuses the genomewide search for autism gene

1
Comparative network analysis of neurological
disorders focuses the genome-wide search for
autism genes.
Dennis P. Wall, PhD Center for Biomedical
Informatics dpwall_at_hms.harvard.edu http//wall.hms
.harvard.edu
2
Outline

• Rationale Biological Significance (30 mins)
• Present status (5 mins)
• Project Plan (25 mins)

3
Introduction

• Polygenic Multigenic
• Many genes have been linked to autism
• Few genes have been replicated in across studies
• Difficult for a single researcher to grasp the
  complexity of the autism gene landscape

4
StatisticsU.S. number of cases 1992-2006
http//www.fightingautism.org
5
Behavioral overlap with other disorders
Schizophrenia
Angelman
Autism
Epilepsy
Fragile X
Seizure Disorder
Rett Syndrome
Mental Retardation
Tuberous Sclerosis
Others??
6
Approach

• Build the network of all genes implicated in
  Autism to date
• Conduct large comparative analysis of Autism and
  other neurological disorders at the level of
  genes, biological processes, and networks
• Leverage existing research on Autism-related
  disorders to find new genetic leads.

7
Building Gene Lists for All Neurological
Disorders (433)
Gene Lists
Ataxia
OMIM
NINDS
Epilepsy
Asperger Fragile X Tourettes OCD
OCD
GeneCards
Autism
Disease source
Disease gene database
Gene-Disease sources
8
Autism Cluster
Genes
1100100101 1110101011 1001010100 1001011101 //
1101011101
Disorders
Autism Cluster
9
Network Construction

• Data derived from STRING (http//string.embl.de/)
• Integration of p-p interaction (interactome),
  co-expression (transcriptome), orthology
  (orthologome),text (bibliome), and other lines of
  evidence.
• Focus on creating a networks of possible
  interactions within a normal cell using
  classification methods (random forests)

10
Correlated Expression
Sequence coEvolution
A
B
P-P Interaction
Random Forest Decision
B
D1
D2
D3
D4
D5
A
D1
D3
1,0,2,1,0
D3
D4
Text (aka Bibliome)
D2
FXYD1 is identified as a MeCP2 target gene whose
de-repression may directly contribute to Rett
syndrome neuronal pathogenesis
D3
D4
D5
D3
D1
Yes
No
D2
D4
http//www.ploscompbiol.org/article/infodoi/10.13
71/journal.pcbi.0030043
11
Networks for all AC disorders
FragileX (97N/100E)
Hypoxia (586 N/4359E)
Microcephaly (135N/166E)
Rett (48N/74E)
Tuberous Sclerosis (110N/204E)
Angelman (51N/57E)
Inf. Hypotonia (29N/16E)
Mental Retardation (573N/1035E)
Hypotonia (154N/208E)
Autism (145N/164E)
Ataxia (428N/1489E)
Spasticity (62N/40E)
Seizure Disorder (35N/13E)
Asperger (15N/9E)
autworks.hms.harvard.edu
12
Multi-disorder component of autism (MDAG)

• 66 out of 127 involved in at least one member of
  the autism cluster
• Highly connected component of the autism network

13
(No Transcript)
14
Significantly enriched MDAG processes
Cell Proliferation
P 2.7E-02
CNS Development
P 3.29E-11
Ion Transport
P 7.68E-10
Synaptic Transmission
P 2.45E-04

• Fishers exact test
• Bonferroni adjustment
• 14648 biological processes from Gene Ontology
  tested

15
Process-Driven Predictions
Putative New Genes
Biological Processes
Autism Cluster Disorders
Fragile X
CNS development
64 new genes, all of which occur in 2 or more of
the Autism Cluster Disorders
Synaptic Transmission
Tuberous Sclerosis
Ion Transport
Cell Proliferation
Seizure Disorder
Mental Retardation
16
Experimental Validation

• GEO6575 (from UC Davis M.I.N.D. institute)
• White blood cell Affymetrix U133plus2.0
• 17 samples of autistic children without
  regression
• 18 children with regression
• 9 children with mental retardation or
  developmental delay
• 12 typically developing children from the general
  population

17
Blood for Brain
18
Autism without regression (17)
Autism with regression (18)
19
Experimental Validation

• GEO6575 (from U.C. Davis M.I.N.D. institute)
• White blood cell Affymetrix U133plus2.0
• 17 samples of autistic patients without
  regression
• 18 patients with regression
• 9 patients with mental retardation or
  developmental delay
• 12 typically developing children from the general
  population

20
Data-driven approach to FDR detection can be
ineffective

• Standard data-driven application of false
  discovery rate control yields few genes below FDR
  threshold of 0.05. (with these data, only 2 genes
  survive)
• This is a frequent circumstance in instances of
  weak signal and large background noise (e.g.
  microarray experiments)

21
Results of process-driven search

• 43 Process-derived gene predictions had
  FDR-adjusted p values lt0.05
• Highly significant rate of validation -- 65 of
  predictions confirmed by expression data

22
Network-Driven Predictions
23
Results of network-driven search

• 267 occurred in 1 autism cluster disorder
• 58 occurred in 2
• 17 in 3
• 3 in 4 sibling disorders
• A total of 345 new predictions

24
Results of network-driven search

• 301 had FDR-adjusted p values lt0.05
• 90 (!) of predictions verified by expression
  data

25
Prior knowledge focuses whole-genomic search

• 43 Process-derived gene predictions had
  FDR-adjusted p values lt0.05. 65
• 301 Network-derived gene predictions had
  FDR-adjusted p values lt0.05. 90

The rate of validation in both cases is
significantly non-random
26
Top 20 genes occurring in 3 or more Autism
Sibling Disorders
For many of these candidates, their roles in
neurological impairment have been studied in
autism cluster disorders, but not in autism.
27
Molecular Triangulation
Mental Retardation Fragile X Hypotonia
Ataxia Hypoxia
AR
SLC16A2
Microcephaly Rett Syndrome Spasticity Tuberous
Sclerosis
L1CAM
OPHN1
FXN
MYO5A
SLC6A8
FLNA
PAFAH1B1
28
Conclusions

• Previous research has implicated between 100 and
  1500 genes as contributors to the molecular
  physiology of Autism.
• Our knowledge-driven approach provides a logical
  means to filter the genome wide search.

29
Conclusions

• Global ask swamped by noisy signal
• Informed, knowledge-driven ask results in
  biologically significant gene predictions
• Comparative analysis of Autism with related
  neurological disorders provides a focused search
  for novel gene candidates

30
Autworks

• Autworks is a web-driven navigation system that
  allows any researcher to view and search through
  the network of genes implicated in autism and
  related neurological disorders
• Built to aid and abet the role of serendipity and
  inspiration for researchers working on autism and
  other complex neuro diseases.
• http//autworks.hms.harvard.edu

31
Autworks now
32
The Plan

• Bring our analytical strategies and Autworks to
  the cloud
• Beef up underbelly using AWS storage and the
  Amazon Turkforce
• Scale up comparative network analysis
• Enlarge validation database, verify/re-verify
  computational predictions, robustify the
  candidates

33
Aim 1 Build the neurological disease gene core
of the Autworks relational database
Can be queried with a disease or gene term
34
Aim 1 Steps

• (1) Extract the entire set of neurological
  disorders listed by NINDS (currently 433) to
  ensure that we can find any and all commonalities
  to Autism.
• (2) Mine all databases in above Table that can be
  searched using a disease term as the query,
  specifically the Online Mendelian Inheritance in
  Man (OMIM), GeneCards, Chromosomal Variation in
  Man, the Human Gene Mutation Database (HGMD), and
  SNPedia.
• (3) Combine and import the features from each of
  the online resources into a relational database
  that will become the backend of Autworks, being
  careful to remove any redundancies.
• (4) Cross-reference resources to comprehensively
  populate data model.

35
Gene-disease data model Gene Core
This data model will share much in common with
Variome projects database
36
MeSH Major Topics
GeneTagger
Candidate gene filtered
MeSH term filtered
PMID 17304222 We identified an important
component for controlled actin assembly, abelson
interacting protein-1 (Abi-1), as a binding
partner for the postsynaptic density (PSD)
protein ProSAP2/Shank3. During early neuronal
development, Abi-1 is localized in neurites and
growth cones at later stages, the protein is
enriched in dendritic spines and PSDs
PMID 17173049 SHANK3 (also known as ProSAP2)
regulates the structural organization of
dendritic spines and is a binding partner of
neuroligins genes encoding neuroligins are
mutated in autism and Asperger syndrome. Here, we
report that a mutation of a single copy of SHANK3
on chromosome 22q13 can result in language and/or
social communication disorders...
Can we Turkify this process???
Annotator Checks Accuracy through BioNotate system
Results Gene-Gene Gene-Disease Corpora
ABI1
Shank3
Shank3
Autism
37
Aim 2 Build interaction network cores for
Autworks
38
Network core
Interaction Core
Ataxia
GO
Co-Ex
Mental Retardation
Can we cloud it up???
Classifier
P-P intx
Bibliome
Phylo-profiles
Autism
39
Aim 3 comparative network analysis on the cloud

• Find disease filtered interacting partners
• Find shortest paths btw candidates
• Find minimal subnetworks
• Verify and reconstruct networks appropriately

Autism
Schizophrenia
40
Mental Retardation
Genetic Landscape of Autism
Rett Syndrome
Angelman Syndrome
41
Autism Diseaseome
42
Acknowledgments

• Zak Kohane
• Matt Huyck
• Tom Monaghan
• Todd DeLuca
• Nieves Mendizabel
• Paco Esteban
• Joaquin Goni

• Alal Eran
• Michal Galdzicki
• Lou Kunkel
• Alexa McCray
• Leon Peshkin