Characteriza*on of Soma*c Muta*ons in Cancer Genomes

Transcription

1 Characteriza*on of Soma*c Muta*ons in Cancer Genomes Ben Raphael Department of Computer Science Center for Computa*onal Molecular Biology

2 Soma*c Muta*ons and Cancer Clonal Theory (Nowell 1976) Passenger muta*ons Founder cell typical tumor : Driver muta*on Time (cell divisions) ~10 driver muta*ons 100 s 1000 s of passenger muta*ons Cell popula*on CGTAATTAG CGTCATTAG Single nucleo*de variant Copy number variants Structural variants

3 Soma*c Muta*ons and Cancer Clonal Theory (Nowell 1976) Passenger muta*ons Founder cell typical tumor : Driver muta*on Time (cell divisions) ~10 driver muta*ons 100 s 1000 s of passenger muta*ons Cell popula*on Sequence genome

4 Challenges in Cancer Genomics 1. Measurement of all soma*c muta*ons. Algorithms Human genome: ~3 billion bp Next- genera6on DNA sequencing CATTCAGTAG GGTAGTTAG TATAATTAG AGCCATTAG GGTAAACTAG CGTACCTAG s million reads Read: bp Genome and Muta6ons

5 Challenges in Cancer Genomics 1. Measurement of all soma*c muta*ons. 1. De novo assembly Algorithms Human genome: ~3 billion bp Next- genera6on DNA sequencing 2. Resequencing CATTCAGTAG GGTAGTTAG TATAATTAG AGCCATTAG GGTAAACTAG CGTACCTAG s million reads Read: bp

6 Challenges in Cancer Genomics 1. Measurement of all soma*c muta*ons. 1. De novo assembly Algorithms Human genome: ~3 billion bp Next- genera6on DNA sequencing 2. Resequencing Right breakpoint (a,b) Geometric Analysis of Structural Varia*on (GASV) Lea breakpoint Bashir, et al. PLOS Comp. Biol. (2008) Sindi et al. Bioinforma3cs/ISMB (2009), Sindi et al. (2011) [In revision] GASVPro

7 Challenges in Cancer Genomics 1. Measurement of all soma*c muta*ons. Algorithms gene Human genome: ~3 billion bp Next- genera6on DNA sequencing Reads of bp Genome and Muta6ons 2. Dis6nguish func*onal (driver) muta*ons from background (passenger) muta*ons.

8 Driver Muta*ons Next- genera*on DNA sequencing gene : soma*c muta*on Recurrent muta*ons/mutated genes à driver muta*ons Dis6nguish func*onal (driver) muta*ons from background (passenger) muta*ons.

9 Recurrent Muta*ons genes pa6ents f Muta6on matrix = mutated = 1 = not mutated = 0 gene : soma*c muta*on Mutated more that expected by chance? Single- gene test à Mul*ple hypotheses correc*on 91 GBM samples. TCGA, Nature (2008) 316 OV samples. TCGA, Nature (2011) Table 2 Significantly mutated genes in HGS-OvCa Gene No. of mutations No. validated No. unvalidated TP BRCA CSMD NF CDK FAT GABRA BRCA RB Validated mutations are those that have been confirmed with an independent assay. Most of them are

10 Mutated Pathways Standard approach: enrichment of muta*ons on known pathways. genes pa6ents p53 pathway: 87% of pa*ents Limita6ons Only exis*ng pathways are tested. Topology of pathways ignored. Pathways are interconnected (crosstalk). TCGA GBM. Nature, 2008.

11 Pathways à Groups " The view that muta*ons affect discrete signaling pathways...has proved to be too simplis*c. [Cancer genes] depend on one another for their selec*ve advantage, and they affect mul1ple pathways that intersect and overlap. Frank McCormick, Signalling networks that cause cancer, Trends in Gene3cs (1999).

12 Advantage of Large Datasets? Prior knowledge of groups of genes Known pathways Interac*on Network None Prior knowledge Number of Hypotheses (< 6 genes) (< 6 genes)

13 Two Algorithms Prior knowledge of groups of genes Known pathways Interac*on Network None Prior knowledge HotNet subnetworks of interac*on network Number of Hypotheses Dendrix Exclusive gene sets

14 HotNet: Problem Defini*on Given: 1. Network G = (V, E) V = genes. E = interac*ons b/w genes 2. Binary muta*on matrix Genes = mutated = not mutated Pa6ents Find: Connected subnetworks mutated in a significant number of pa*ents mutated in pa*ent if 1 gene mutated in pa*ent

15 (Local) Topology Maoers Single path between mutated genes Path between mutated genes is one of many through node. Example: TP53 has 238 neighbors in HPRD network

16 Heat Diffusion on Graph

17 Mutated subnetworks: HotNet* Muta6on Matrix Genes Human Interac6on Network = mutated genes Pa*ents (1) Muta6on heat diffusion Extract significantly hot subnetworks Hot (2) Cold *F. Vandin, E. Upfal, and B. J. Raphael. J. Comp.Biol. (2011). Also RECOMB (2010).

18 Sta*s*cal Test Muta6on Matrix Genes Random Binary Matrix Genes Pa*ents Pa*ents X s = number of subnetworks s genes Two- stage mul6- hypothesis test: Rigorously bound FDR.

19 HPRD network nodes edges Results: TCGA Ovarian ~20000 Genes 316 Pa*ents Muta*on & copy number HotNet 27 subnetworks with 7 genes (P < 0.03) TCGA. Nature (2011)

20 Ovarian Subnetworks TCGA. Nature (2011)

21 Ovarian Subnetworks Kegg Pathway Notch signaling (p < 6x10-7 ) 12/27 subnetworks significantly overlap known pathways (KEGG) or protein complexes (PINdb) TCGA. Nature (2011)

22 Ovarian Subnetworks Known Protein Complexes Cohesin (P < 5x10-8 ) Kegg Pathway Cell cycle (P < 2x10-5 ) TCGA. Nature (2011)

23 Results: Brain Cancer 91 Samples HPRD network nodes edges 601 Genes Muta*on & copy number HotNet 2 subnetworks with 20 genes (P < 0.01)

24 Brain: Muta*ons + Copy number RTK/RAS/PI(3)K RB1 p53 Subnetwork Enrichment p- val s RTK/RAS/PI(3)K P53 RB x10-6 4x

25 Two Algorithms Prior knowledge of groups of genes Known pathways Interac*on Network None Prior knowledge HotNet subnetworks of interac*on network Number of Hypotheses Dendrix Exclusive gene sets

26 Pathways and Muta*ons Driver muta3ons are rare. Cancer pathway has one driver muta*on (gene) per pa*ent [Vogelstein and K. W. Kinzler (2004), Yeang, McCormick, and Levine (2008)] 1. Exclusivity Genes EGFR RAF cell membrane RAS MEK Pa6ents MAPK Transcrip*on factors [Thomas, et al. (2007)]

27 Pathways and Muta*ons Driver muta3ons are rare. Cancer pathway has one driver muta*on (gene) per pa*ent [Vogelstein and K. W. Kinzler (2004), Yeang, McCormick, and Levine (2008)] 1. Exclusivity EGFR RAF cell membrane RAS Many pa*ents have muta*on in important cancer pathway. 2. Coverage MAPK MEK Transcrip*on factors

28 De novo driver exclusivity (Dendrix*) Given: Binary muta*on matrix A Find: Set M of genes with: Maximum Coverage: maximum number of pa*ents have 1 muta*on in M Mutual Exclusivity: all pa*ents have 1 muta*on in M pa6ents ; genes Difficult computa*onal problem (NP- hard) *Vandin, Upfal, & Raphael. Genome Res. (Advance online) Also RECOMB 2011.

29 De novo driver exclusivity (Dendrix*) Given: Binary muta*on matrix A Find: Set M of genes with: High Coverage: many pa*ents have a 1 muta*on in M Approximate Exclusivity: most pa*ents have 1 muta*on in M Finding largest M is difficult! Greedy algorithm and MCMC algorithm Theore*cal results on convergence and op*mality. pa6ents ; genes Dendrix++: extension with scoring based on probabilis*c model *Vandin, Upfal, & Raphael. Genome Res. (Advance online) Also RECOMB 2011.

30 Experimental Results Cancer data 1. Known muta6ons in mul6ple cancer types [Thomas et al. (2007)] 238 known muta*ons in 1000 pa*ents of 17 different cancer types 2. Lung Adenocarcinoma [Ding et al. (2008)] 623 sequenced genes in 188 pa*ents 3. Brain cancer (GBM) [TCGA (2008)] 601 sequenced genes in 84 pa*ents Array copy number data on all genes 4. Acute Myleloid Leukemia (AML) [TCGA] Whole exome sequencing on ~200 pa*ents 5. Breast Cancer (BRCA) [TCGA] Whole exome sequencing Array copy number on ~400 pa*ents

31 Known muta*ons in mul*ple cancer types 238 known muta*ons tumors from 17 cancer types [Thomas, et al. (2007)] Muta*on: Exclusive Co- occurring None 8 muta*ons cover 94% of mutated tumors Only 15 tumors with >1 muta*on (p < 0.01)

32 Lung Adenocarcinoma 623 sequenced genes in 188 pa*ents [Ding et al. (2008)] (p < 0.01) (p < 0.01)

33 Brain Cancer (GBM) Muta*ons (single- nucleo*de and copy number) for 601 genes in 84 samples [TCGA (2008)] (p < 0.01) (p < 0.01) (p < 0.01)

34 Challenges in Cancer Genomics 1. Measurement of all soma*c muta*ons. Algorithms gene Human genome: ~3 billion bp Next- genera6on DNA sequencing Reads of bp Genome and Muta6ons 2. Dis6nguish func*onal (driver) muta*ons from background (passenger) muta*ons.

35 Summary Prior knowledge of groups of genes Known pathways Prior knowledge Interac*on Network None HotNet subnetworks of interac*on network Number of Hypotheses Dendrix Exclusive gene sets Future: Incorporate more data types (methyla*on, gene expression). Perform pre/post filtering of predic*ons.

36 Acknowledgements Fabio Vandin Eli Upfal Hsin- Ta Wu Edward Rice Tim Ley Li Ding Elaine Mardis Rick Wilson and others Funding

37 Shameless Adver*sing 1) Soaware available. 2) Student* and postdoc posi6ons available. * Ph.D. programs in Computer Science, Computa*onal Biology, and Molecular Biology. Web: hop://cs.brown.edu/people/braphael