Pan-cancer network analysis identifies combinations of rare somatic mutations across pathways and protein complexes

Size: px

Start display at page:

Download "Pan-cancer network analysis identifies combinations of rare somatic mutations across pathways and protein complexes"

Pierce Sherman
6 years ago
Views:

Pan-cancer network analysis identifies combinations of rare somatic mutations across pathways and protein complexes Nature Genetics 47, 106-114

1 Pan-cancer network analysis identifies combinations of rare somatic mutations across pathways and protein complexes Nature Genetics 47, (2015) doi:101038/ng3168 Max Leiserson RECOMB 2015 April 14, 2015 Raphael Lab Department of Computer Science & Center for Computational Molecular Biology

2 Identifying cancer driver genes Cancer Genome Landscapes >999% of mutations are passengers Vogelstein et al (2013) 3-8 drivers per tumor" 2

3 Identifying cancer driver genes Cancer Genome Landscapes Vogelstein et al (2013) >999% of mutations are passengers 3-8 drivers per tumor" Compare variation across tumors Tumor 1 Tumor 2 Tumor N = gene = mutation Single nucleotide variants Copy number aberrations Gene expression 2

4 Identifying cancer driver genes Cancer Genome Landscapes Vogelstein et al (2013) >999% of mutations are passengers 3-8 drivers per tumor" Compare variation across tumors = gene Tumor 1 Tumor 2 Tumor N = mutation Single nucleotide variants Copy number aberrations Gene expression Significance Score Identify cancer driver genes genes Gene Mutations weighted by: Recurrence Gene length Mutation context Expression level Replication timing N= samples long tail 2

5 Cancer driver mutations target pathways Driver mutations confer a growth advantage to the tumor driver genes are members of cancer signaling pathways 3

6 Cancer driver mutations target pathways Driver mutations confer a growth advantage to the tumor driver genes are members of cancer signaling pathways STAT PI3K MAPK TGF- RAS DNA damage control Transcriptional regulation Cell cycle/ apoptosis Selective growth advantage NOTCH Chromatin modification HH APC Vogelstein et al (Science 2013) Genome Maintenance 3

7 Cancer driver mutations target pathways Driver mutations confer a growth advantage to the tumor driver genes are members of cancer signaling pathways STAT PI3K MAPK TGF- RAS DNA damage control Transcriptional regulation Cell cycle/ apoptosis Selective growth advantage NOTCH Chromatin modification HH APC Vogelstein et al (Science 2013) Genome Maintenance Significance Score genes A B N= samples A B interacts with 3

Cancer driver mutations target pathways Driver mutations confer a growth advantage to the tumor driver genes are members of cancer signaling pathways STAT PI3K MAPK TGF- RAS DNA damage control

8 Cancer driver mutations target pathways Driver mutations confer a growth advantage to the tumor driver genes are members of cancer signaling pathways STAT PI3K MAPK TGF- RAS DNA damage control Transcriptional regulation Cell cycle/ apoptosis Selective growth advantage NOTCH Chromatin modification HH APC Vogelstein et al (Science 2013) Genome Maintenance Significance Score genes A B N= samples A B interacts with TCGA Ovarian (Nature 2011) 3

9 Testing known gene sets and pathways Input data Mutation data Tumor 1 Tumor 2 Tumor N (eg most mutated genes: EGFR, KRAS, BRAF) Gene set database PI3K RAS Cell cycle/ apoptosis NOTCH STAT MAPK TGF- Selective growth advantage DNA damage control Transcriptional regulation Chromatin modification HH APC Genome Maintenance 4

10 Testing known gene sets and pathways Input data Mutation data Tumor 1 Tumor 2 Tumor N (eg most mutated genes: EGFR, KRAS, BRAF) Enrichment tests GSEA [1,2] DAVID [3,4] Gene set database PI3K RAS Cell cycle/ apoptosis NOTCH STAT MAPK TGF- Selective growth advantage DNA damage control Transcriptional regulation Chromatin modification HH APC Genome Maintenance [1] Mootha et al Nat Genet (2003) [3] Huang et al Nat Protoc (2009) [2] Subramanian et al PNAS (2005) [4] Huang et al Nucleic Acids Res (2009) 4

11 Testing known gene sets and pathways Input data Mutation data Tumor 1 Tumor 2 Tumor N (eg most mutated genes: EGFR, KRAS, BRAF) Enrichment tests GSEA [1,2] DAVID [3,4] Enriched gene sets MYC ARID1A FBXW7 APC SOX9 CTNNB1 Gene set database PI3K RAS Cell cycle/ apoptosis NOTCH STAT MAPK TGF- Selective growth advantage DNA damage control Transcriptional regulation Chromatin modification HH APC Genome Maintenance [1] Mootha et al Nat Genet (2003) [3] Huang et al Nat Protoc (2009) [2] Subramanian et al PNAS (2005) [4] Huang et al Nucleic Acids Res (2009) 4

12 Testing known gene sets and pathways Input data Mutation data Tumor 1 Tumor 2 Tumor N (eg most mutated genes: EGFR, KRAS, BRAF) Enrichment tests GSEA [1,2] DAVID [3,4] Enriched gene sets MYC ARID1A FBXW7 APC SOX9 CTNNB1 Gene set database STAT PI3K MAPK TGF- RAS DNA damage control Genome Maintenance Transcriptional regulation Cell cycle/ apoptosis Selective growth advantage NOTCH Chromatin modification HH APC Key drawbacks Novel pathways and crosstalk? Topology of interactions? Handling large and/or overlapping pathways? [1] Mootha et al Nat Genet (2003) [2] Subramanian et al PNAS (2005) [3] Huang et al Nat Protoc (2009) [4] Huang et al Nucleic Acids Res (2009) 4

13 Significantly mutated subnetworks of a protein-protein interaction network Protein-protein interaction networks Nodes: genes/protein Edges: connect genes if the proteins they encode physically interact Unweighted, undirected Goal: identify connected subnetworks with more mutations than expected by chance No Mutations 5

14 Significantly mutated subnetworks of a protein-protein interaction network Protein-protein interaction networks Nodes: genes/protein Edges: connect genes if the proteins they encode physically interact Unweighted, undirected Goal: identify connected subnetworks with more mutations than expected by chance No Mutations ~10 18 subnetworks of size k=5 5

15 Significantly mutated subnetworks of a protein-protein interaction network Protein-protein interaction networks Nodes: genes/protein Edges: connect genes if the proteins they encode physically interact Unweighted, undirected Network Nodes Edges Diameter ASP HPRD 9,205 36, HINT+HI2012 9,859 40, irefindex 12,129 91, MultiNet 14, , Goal: identify connected subnetworks with more mutations than expected by chance Low diameter Most genes have a high-scoring neighbor No Mutations ~10 18 subnetworks of size k=5 5

16 Significantly mutated subnetworks of a protein-protein interaction network Protein-protein interaction networks Nodes: genes/protein Edges: connect genes if the proteins they encode physically interact Unweighted, undirected Network Nodes Edges Diameter ASP HPRD 9,205 36, HINT+HI2012 9,859 40, irefindex 12,129 91, MultiNet 14, , Goal: identify connected subnetworks with more mutations than expected by chance Low diameter Most genes have a high-scoring neighbor Must analyze mutations and local topology simultaneously! No Mutations ~10 18 subnetworks of size k=5 5

17 Outline 1 A new algorithm, HotNet2 2 Application to TCGA Pan-Cancer data A101D A101* A101E L106T P179V E187G R252L V276T 3 Comparison of HotNet2 to other methods 6

18 Encoding mutations and graph topology with heat diffusion Heat Heat diffusion process t=0 t=1 t=n u u u Place heat on node u Heat diffuses from node u to u's neighbors Related to random walks and network propagation 7

19 Encoding mutations and graph topology with heat diffusion Heat Heat diffusion process t=0 t=1 t=n u u u Place heat on node u Heat diffuses from node u to u's neighbors Related to random walks and network propagation HotNet (Vandin et al JCB & RECOMB 2010) Mutations = heat sources Heat diffusion Extract hot subnetworks = mutated gene 7

20 HotNet applied to TCGA data TCGA Papers (~300 samples) Leukemia (NEJM 2013) Kidney (Nature 2011) Ovarian (Nature 2011) HotNet (Vandin et al JCB & RECOMB 2010) Mutations = heat sources Heat diffusion Extract hot subnetworks = mutated gene 8

21 HotNet applied to TCGA data TCGA Pan-Cancer (>3000 samples) TCGA Papers (~300 samples) Leukemia (NEJM 2013) Kidney (Nature 2011) Ovarian (Nature 2011) u HotNet finds many star subnetworks with one central, hot node HotNet (Vandin et al JCB & RECOMB 2010) Mutations = heat sources Heat diffusion Extract hot subnetworks = mutated gene 8

22 HotNet Algorithm Input Output h1 hn A = adjacency matrix h = gene scores Connected components Threshold at δ Heat kernel f(a, t) h1 0 0 hn = s11 sn1 s1n snn symmetrize r11 rn1 r1n rnn Diffusion matrix (symmetric) Time parameter t Similarity matrix (asymmetric) sij = heat on vertex i at time t given initial heat hj on vertex j at time 0 9

23 Direction of heat is important HotNet can fail HotNet s heat is symmetric Potential artifacts u sends the same heat to v even though u has much higher degree v u u Hot nodes with high degree often form large star subnetworks with many cold nodes Star graph Heat kernel f(a, t) h1 0 0 hn = s11 sn1 s1n snn symmetrize r11 rn1 r1n rnn Diffusion matrix (symmetric) Similarity matrix S (asymmetric) sij = heat on vertex i at time t given initial heat hj on vertex j at time 0 10

24 HotNet2 algorithm (HotNet diffusion oriented subnetworks) Need to consider the source of heat Encode directionality with asymmetric heat diffusion Hot genes do not necessarily implicate their neighbors Hot subnetworks have a directed path between each pair of nodes sij = heat on vertex i at equilibrium given initial heat hj on vertex j at time 0 Similarity Identify strongly connected components Similarity matrix S Asymmetric! similarity of u and v Leiserson, Vandin et al Nat Genet (2015) 11

25 HotNet2 algorithm (HotNet diffusion oriented subnetworks) Need to consider the source of heat Encode directionality with asymmetric heat diffusion Hot genes do not necessarily implicate their neighbors Hot subnetworks have a directed path between each pair of nodes sij = heat on vertex i at equilibrium given initial heat hj on vertex j at time 0 Similarity Identify strongly connected components Similarity matrix S Asymmetric! Leiserson, Vandin et al Nat Genet (2015) similarity of u and v TCGA Papers Gastric, Nature (2014) Thyroid, Cell (2014) 11

26 HotNet2 vs HotNet Input HotNet2 A = adjacency matrix h1 hn h = gene scores Strongly connected components Threshold at δ HotNet: Heat kernel f(a, t) HotNet2: Insulated heat diffusion Diffusion matrix (HotNet: symmetric HotNet2: asymmetric) h1 0 0 hn = s11 sn1 s1n snn Similarity matrix (asymmetric) symmetrize Connected components r11 rn1 r1n rnn Threshold at δ HotNet 12

27 Statistical test Evaluate graph partition with rigorously bounded False Discovery Rate (FDR) Gene scores Randomized gene scores x2 = 3 x3 = 1 x2 = 1 x3 = 0 Leiserson, Vandin et al Nat Genet (2015) Xk: number of subnetworks of size k Pr(Xk xk h, δ) 13

28 Outline 1 A new algorithm, HotNet2 2 Application to TCGA Pan-Cancer data A101D A101* A101E L106T P179V E187G R252L V276T 3 Comparison of HotNet2 to other methods 14

29 TCGA Pan-Cancer Cancer Tumor samples BLCA 99 BRCA 772 COAD/READ 224 GBM 291 HNSC 306 KIRC 417 LAML 196 LUAD 230 LUSC 178 OV 316 UCEC 248 Samples Color 3,110 tumors of 12 cancer types Weinstein et al Nature Genetics (2013) Single nucleotide variants A101D A101* A101E L106T Number of CNAs P179V E187G CCND1 MYC ERBB2 EGFR CDKN2A R252L KRAS MLL3 APC VHL Mutations V276T PTEN 11,565 mutated, expressed genes PIK3CA SNVs and CNAs in 3,110 samples among 11,565 expressed genes Hot Cold ,000 1,100 1,200 1,300 Number of SNVs Copy number aberrations Patient Chromosome arm Hot Cold TP53 15

30 HotNet2 runs on TCGA Pan-Cancer dataset Mutation & copy number data A101D A101* A101E L106T P179V E187G R252L V276T Interaction network Significantly mutated subnetworks Patient HotNet2 Chromosome arm 11,565 mutated genes in 3,110 samples 16

31 HotNet2 runs on TCGA Pan-Cancer dataset Mutation & copy number data A101D A101* A101E L106T P179V E187G R252L V276T Interaction network Significantly mutated subnetworks Patient HotNet2 Chromosome arm 11,565 mutated genes in 3,110 samples HINT+HI2012 (P < 001) irefindex 90 (P < 001) Multinet (P < 001) 40,704 interactions 9,858 proteins 91,808 interactions 12,128 proteins 109,569 interactions 14,398 proteins 16

HotNet2 runs on TCGA Pan-Cancer dataset Mutation & copy number data A101D A101* A101E L106T P179V E187G R252L V276T Interaction network Significantly mutated subnetworks Patient HotNet2 Chromosome

32 HotNet2 runs on TCGA Pan-Cancer dataset Mutation & copy number data A101D A101* A101E L106T P179V E187G R252L V276T Interaction network Significantly mutated subnetworks Patient HotNet2 Chromosome arm 11,565 mutated genes in 3,110 samples HINT+HI2012 (P < 001) irefindex 90 (P < 001) Multinet (P < 001) 40,704 interactions 9,858 proteins 91,808 interactions 12,128 proteins 109,569 interactions 14,398 proteins Consensus subnetworks 16 consensus subnetworks with 4 genes (P=0004) 13 linkers between consensus subnetworks 16

33 HotNet2 Consensus HotNet2 Runs HINT+HI2012 (P < 001) irefindex 90 (P < 001) Multinet (P < 001) 1 Interaction networks Many low-confidence edges 075 TPR = Sensitivity High confidence edges High- and some lowconfidence edges FPR = 1-Specificity

34 HotNet2 Consensus HotNet2 Runs HINT+HI2012 (P < 001) irefindex 90 (P < 001) Multinet (P < 001) TPR = Sensitivity Interaction networks Multinet (110K edges)? irefindex (90K edges)?? HINT+HI2012 (40K edges) FPR = 1-Specificity

HotNet2 Consensus HotNet2 Runs HINT+HI2012 (P < 001) irefindex 90 (P < 001) Multinet (P < 001) Consensus 16 consensus subnetworks 13 linkers between subnetworks 1 075 TPR = Sensitivity 05 025

35 HotNet2 Consensus HotNet2 Runs HINT+HI2012 (P < 001) irefindex 90 (P < 001) Multinet (P < 001) Consensus 16 consensus subnetworks 13 linkers between subnetworks TPR = Sensitivity Interaction networks Multinet (110K edges)?? irefindex (90K edges) HINT+HI2012 (40K edges)? 3 Main Idea: Incorporate lowconfidence edges but give highconfidence edges more weight Consensus Linkers Consensus FPR = 1-Specificity Consensus Graph Edges connect genes identified by HotNet2 in the same subnetwork 17

36 HotNet2 Consensus Subnetworks GBM LAML Frequently and rarely mutated cancer genes OV ErbB signaling (179%) EGFR ERBB2 ERBB4 **OSMR **LRIG1 *ELF3 **AREG Core-binding factors(27%) RUNX1 CBFB **ELF4 BRCA P53 signaling(684%) TP53 CDKN2A MDM4 CCND1 PTEN NPM1 39 Others Linkers (212%) TP53INP1 WT1 **CCDC88A **RNF20 **CABLES1 *NOTCH2 PI3K signaling(326%) PIK3CA NRAS KRAS HNSC LUSC NOTCH signaling(198%) MHC Class I (312%) *HLA-A NOTCH1 **HLA-B **P4HTM B2M **CD1D *NOTCH3 **NOTCH4 **JAG1 CCNE1 **MAML111 Others CLASP and CLIP (2%) **CLIP2 **CLASP2 **CLASP1 ASCOM complex (1691%) ASXL1 MLL3 **FOXK2 MLL2 **N4BP2 KDM6A **PROSER1 **KDM1B *E2F3 BAP1 complex (71%) BAP1 **ASXL2 **ANKRD17 *FOXK1 Cohesin complex (736%) PIK3R1 MAP2K4 MAP3K1 HRAS *RASA1 ATM STK11 (19%) **BOC **CDON BRAF 8 Others SWI/SNF complex (168%) ARID1A PBRM1 SMARCA4 ARID2 *ARID1B **ADNP SMARCB1 Condensin Complex (42%) *SMC4 **NCAPG2 **NCAPD3 **NCAPH2 **NCAPD2 **SMC2 (07%) SPOP MYD88 LUAD UCEC STAG2 **PDS5A **STAG1 *SMC1A *RAD21 KEAP1, NFE2L2 (849%) KEAP1 NFE2L2 **CHD6 **WAPAL **PDS5B **PTMA **WAC BLCA COADREAD KIRC 18

37 HotNet2 Consensus Subnetworks HNSC LUSC OV *HLA-A **HLA-B B2M **P4HTM **CD1D GBM NOTCH signaling(198%) MHC Class I (312%) CLASP and CLIP (2%) **CLIP2 **CLASP2 **CLASP1 ASCOM complex (1691%) ASXL1 MLL3 **FOXK2 MLL2 **N4BP2 KDM6A **PROSER1 **KDM1B *E2F3 P53 signaling(684%) ErbB signaling (179%) EGFR ERBB2 ERBB4 **OSMR **LRIG1 BAP1 complex (71%) BAP1 **ASXL2 **ANKRD17 *FOXK1 Cohesin complex (736%) STAG2 **PDS5A **STAG1 *SMC1A *RAD21 LAML *ELF3 **AREG CBFB Linkers (212%) TP53INP1 WT1 TP53 CDKN2A MDM4 **CCDC88A CCND1 PTEN **RNF20 NPM1 39 Others **CABLES1 *NOTCH2 PIK3R1 MAP2K4 Core-binding factors(27%) (19%) **BOC **CDON KEAP1, NFE2L2 (849%) KEAP1 NFE2L2 **CHD6 RUNX1 **ELF4 PI3K signaling(326%) PIK3CA NRAS KRAS BRAF 8 Others PBRM1 SMARCA4 ARID2 *ARID1B **ADNP SMARCB1 BRCA (07%) MAP3K1 NOTCH1 HRAS SPOP MYD88 *NOTCH3 **JAG1 *RASA1 **NOTCH4 CCNE1 ATM SWI/SNF complex (168%) **MAML111 Others STK11 ARID1A Condensin Complex (42%) *SMC4 **NCAPG2 **NCAPD3 **NCAPH2 **NCAPD2 **SMC2 LUAD UCEC Frequently and rarely mutated cancer genes Well-known cancer pathways PI(3)K signaling p53 signaling NOTCH signaling ErbB signaling Linkers: HRAS, STK11, ATM **WAPAL **PDS5B **PTMA **WAC BLCA COADREAD KIRC 18

38 HotNet2 Consensus Subnetworks HNSC LUSC OV GBM NOTCH signaling(198%) MHC Class I (312%) **HLA-B **P4HTM *HLA-A B2M **CD1D CLASP and CLIP (2%) **CLASP2 **CLIP2 MLL2 **CLASP1 ASCOM complex (1691%) **N4BP2 MLL3 *E2F3 KDM6A **PROSER1 ErbB signaling (179%) Cohesin complex (736%) STAG2 **PDS5A **STAG1 *SMC1A **OSMR P53 signaling(684%) CDKN2A CCND1 NPM1 *NOTCH3 TP53 **NOTCH4 NOTCH1 MDM4 PTEN 39 Others **JAG1 CCNE1 **MAML111 Others BAP1 complex (71%) ASXL1 **FOXK2 ERBB2 BAP1 **KDM1B EGFR ERBB4 **LRIG1 *ELF3 **AREG **ASXL2 **ANKRD17 *FOXK1 *RAD21 Linkers (212%) TP53INP1 WT1 **CCDC88A **RNF20 **CABLES1 *NOTCH2 PIK3R1 MAP2K4 MAP3K1 HRAS *RASA1 ATM STK11 **BOC LAML Core-binding factors(27%) (19%) KEAP1, NFE2L2 (849%) NFE2L2 CBFB **CDON KEAP1 **CHD6 RUNX1 **ELF4 PI3K signaling(326%) NRAS BRAF SWI/SNF complex (168%) PBRM1 ARID2 PIK3CA ARID1A KRAS 8 Others SMARCA4 *ARID1B **ADNP SMARCB1 BRCA SPOP Condensin Complex (42%) *SMC4 **NCAPG2 **NCAPD3 **NCAPH2 **NCAPD2 **SMC2 (07%) MYD88 LUAD UCEC Frequently and rarely mutated cancer genes Well-known cancer pathways PI(3)K signaling p53 signaling NOTCH signaling ErbB signaling Linkers: HRAS, STK11, ATM p53 Signaling CUL9 CHD8 NPM1 TP53 PTEN MDM4 CCND1 CDKN2A EPHA3 plus ~30 others Mutations Multinet irefindex HINT+HI2012 **WAPAL **PDS5B **PTMA **WAC BLCA COADREAD KIRC 18

39 HotNet2 Consensus Subnetworks HNSC LUSC OV **HLA-B **P4HTM *HLA-A B2M **CD1D BLCA GBM NOTCH signaling(198%) MHC Class I (312%) CLASP and CLIP (2%) **CLASP2 ASCOM complex (1691%) MLL2 **N4BP2 MLL3 *E2F3 KDM6A **PROSER1 BAP1 complex (71%) ASXL1 **FOXK2 BAP1 **KDM1B **ASXL2 **ANKRD17 *FOXK1 Cohesin complex (736%) STAG2 **PDS5A **STAG1 KIRC LAML KEAP1, NFE2L2 (849%) *SMC1A NFE2L2 **CHD6 *RAD21 **WAPAL **PDS5B **PTMA **WAC CHD8 (mutated in 54 of 3110 samples) Legend **CLIP2 Legend Cancer types **CLASP1 ErbB signaling (179%) ERBB2 **OSMR P53 signaling(684%) CDKN2A CCND1 NPM1 *NOTCH3 TP53 **NOTCH4 NOTCH1 MDM4 PTEN 39 Others **JAG1 CCNE1 **MAML111 Others EGFR ERBB4 **LRIG1 *ELF3 **AREG Linkers (212%) TP53INP1 **CCDC88A **RNF20 **CABLES1 *NOTCH2 PIK3R1 MAP2K4 MAP3K1 HRAS *RASA1 ATM STK11 WT1 **BOC Core-binding factors(27%) (19%) CBFB **CDON KEAP1 RUNX1 PI3K signaling(326%) SWI/SNF complex (168%) COADREAD **ELF4 NRAS BRAF PBRM1 ARID2 PIK3CA ARID1A KRAS 8 Others SMARCA4 *ARID1B **ADNP SMARCB1 *SMC4 **NCAPG2 **NCAPD3 **NCAPH2 **NCAPD2 **SMC2 BRCA SPOP Condensin Complex (42%) (07%) MYD88 BLCA BRCA COADREAD GBM HNSC KIRC LAML LUAD LUSC OV UCEC LUAD UCEC Frequently and rarely mutated cancer genes Well-known cancer pathways PI(3)K signaling p53 signaling NOTCH signaling ErbB signaling Linkers: HRAS, STK11, ATM p53 Signaling CUL9 CHD8 NPM1 TP53 Mutation types PTEN MDM4 CCND1 CDKN2A EPHA3 plus ~30 others Inactivating SNV Amplification Mutations Multinet irefindex HINT+HI2012 SNV Deletion 18

40 HotNet2 Consensus Subnetworks HNSC LUSC OV *HLA-A **HLA-B B2M **P4HTM **CD1D GBM NOTCH signaling(198%) MHC Class I (312%) CLASP and CLIP (2%) **CLIP2 **CLASP2 **CLASP1 ASCOM complex (1691%) ASXL1 MLL3 **FOXK2 MLL2 **N4BP2 KDM6A **PROSER1 **KDM1B *E2F3 P53 signaling(684%) ErbB signaling (179%) EGFR ERBB2 ERBB4 **OSMR **LRIG1 BAP1 complex (71%) BAP1 **ASXL2 **ANKRD17 *FOXK1 Cohesin complex (736%) STAG2 **PDS5A **STAG1 *SMC1A *RAD21 LAML *ELF3 **AREG CBFB Linkers (212%) TP53INP1 WT1 TP53 CDKN2A MDM4 **CCDC88A CCND1 PTEN **RNF20 NPM1 39 Others **CABLES1 *NOTCH2 PIK3R1 MAP2K4 Core-binding factors(27%) (19%) **BOC **CDON KEAP1, NFE2L2 (849%) KEAP1 NFE2L2 **CHD6 RUNX1 **ELF4 PI3K signaling(326%) PIK3CA NRAS KRAS BRAF 8 Others PBRM1 SMARCA4 ARID2 *ARID1B **ADNP SMARCB1 BRCA (07%) MAP3K1 NOTCH1 HRAS SPOP MYD88 *NOTCH3 **JAG1 *RASA1 **NOTCH4 CCNE1 ATM SWI/SNF complex (168%) **MAML111 Others STK11 ARID1A Condensin Complex (42%) *SMC4 **NCAPG2 **NCAPD3 **NCAPH2 **NCAPD2 **SMC2 LUAD UCEC Frequently and rarely mutated cancer genes Well-known cancer pathways PI(3)K signaling p53 signaling NOTCH signaling ErbB signaling Linkers: HRAS, STK11, ATM Recently characterized complexes: SWI/SNF complex ASCOM complex BAP1 complex **WAPAL **PDS5B **PTMA **WAC BLCA COADREAD KIRC 18

41 HotNet2 Consensus Subnetworks HNSC LUSC OV *HLA-A **HLA-B B2M **P4HTM **CD1D BLCA GBM NOTCH signaling(198%) MHC Class I (312%) CLASP and CLIP (2%) **CLIP2 **CLASP2 **CLASP1 ASCOM complex (1691%) ASXL1 MLL3 **FOXK2 MLL2 **N4BP2 KDM6A **PROSER1 **KDM1B *E2F3 P53 signaling(684%) ErbB signaling (179%) EGFR ERBB2 ERBB4 **OSMR **LRIG1 BAP1 complex (71%) BAP1 **ASXL2 **ANKRD17 *FOXK1 Cohesin complex (736%) STAG2 **PDS5A **STAG1 *SMC1A *RAD21 **WAPAL **PDS5B LAML *ELF3 **AREG CBFB Linkers (212%) TP53INP1 WT1 TP53 CDKN2A MDM4 **CCDC88A CCND1 PTEN **RNF20 NPM1 39 Others **CABLES1 *NOTCH2 PIK3R1 MAP2K4 Core-binding factors(27%) (19%) **BOC **CDON KEAP1, NFE2L2 (849%) KEAP1 NFE2L2 **CHD6 **PTMA **WAC RUNX1 **ELF4 PI3K signaling(326%) COADREAD PIK3CA NRAS KRAS BRAF 8 Others PBRM1 SMARCA4 ARID2 *ARID1B **ADNP SMARCB1 BRCA (07%) MAP3K1 NOTCH1 HRAS SPOP MYD88 *NOTCH3 **JAG1 *RASA1 **NOTCH4 CCNE1 ATM SWI/SNF complex (168%) **MAML111 Others STK11 ARID1A Condensin Complex (42%) *SMC4 **NCAPG2 **NCAPD3 **NCAPH2 **NCAPD2 **SMC2 LUAD UCEC Frequently and rarely mutated cancer genes Well-known cancer pathways PI(3)K signaling p53 signaling NOTCH signaling ErbB signaling Linkers: HRAS, STK11, ATM Recently characterized complexes: SWI/SNF complex ASCOM complex BAP1 complex Potentially novel complexes: Cohesin complex Condensin complex MHC Class I proteins KIRC 18

42 HotNet2 Consensus Subnetworks HNSC LUSC OV *HLA-A **HLA-B B2M **P4HTM **CD1D BLCA GBM NOTCH signaling(198%) MHC Class I (312%) CLASP and CLIP (2%) **CLIP2 **CLASP2 **CLASP1 ASCOM complex (1691%) ASXL1 MLL3 **FOXK2 MLL2 **N4BP2 KDM6A **PROSER1 **KDM1B *E2F3 P53 signaling(684%) ErbB signaling (179%) EGFR ERBB2 ERBB4 **OSMR **LRIG1 BAP1 complex (71%) BAP1 **ASXL2 **ANKRD17 *FOXK1 Cohesin complex (736%) STAG2 **PDS5A **STAG1 *SMC1A *RAD21 **WAPAL **PDS5B LAML *ELF3 **AREG CBFB Linkers (212%) TP53INP1 WT1 TP53 CDKN2A MDM4 **CCDC88A CCND1 PTEN **RNF20 NPM1 39 Others **CABLES1 *NOTCH2 PIK3R1 MAP2K4 Core-binding factors(27%) (19%) **BOC **CDON KEAP1, NFE2L2 (849%) KEAP1 NFE2L2 **CHD6 **PTMA **WAC RUNX1 **ELF4 PI3K signaling(326%) COADREAD PIK3CA NRAS KRAS BRAF 8 Others PBRM1 SMARCA4 ARID2 *ARID1B **ADNP SMARCB1 BRCA (07%) MAP3K1 NOTCH1 HRAS SPOP MYD88 *NOTCH3 **JAG1 *RASA1 **NOTCH4 CCNE1 ATM SWI/SNF complex (168%) **MAML111 Others STK11 ARID1A Condensin Complex (42%) *SMC4 **NCAPG2 **NCAPD3 **NCAPH2 **NCAPD2 **SMC2 LUAD UCEC Frequently and rarely mutated cancer genes Well-known cancer pathways PI(3)K signaling p53 signaling NOTCH signaling ErbB signaling Linkers: HRAS, STK11, ATM Recently characterized complexes: SWI/SNF complex ASCOM complex BAP1 complex Potentially novel complexes: Cohesin complex Condensin complex MHC Class I proteins KIRC 18

43 SWI/SNF complex a SWI / SNF Complex Coverage: 168% (523 / 3110 samples) = 5 samples ARID1A (194) PBRM1 (169) SMARCA4 (67) ARID2 (56) *ARID1B (41) **ADNP (21) SMARCB1 (19) SMARCA4 LUAD: 8x10-3 ARID1A UCEC: 158x10-15 BLCA: 488x10-9 PBRM1 KIRC: 7x10-97 S10T N80Y E226Q K231R I245F I252R P427S N453S F456S R522L M523R N528I Y580C E583K M586K M586I M586T N601K L614V L618H G626V I709T Y718C M731K R760H E860K R876L P1048L P1048R R1088Q V1136L T1202K G1231R C1233W E1248G E1287D I1309T D1325N R1411C W1417G M1539I G1554R G1577C L1611P R1670P ARID1B ADNP ARID2 SMARCB1 BLCA: 001 All PPI networks MultiNet irefindex HINT HI missense Scale (AA) PBRM1 BAH_dom Bromodomain HMG_superfamily Legend Legend Cancer types BLCA BRCA COADREAD GBM HNSC KIRC LAML LUAD LUSC OV UCEC Mutation types Inactivating SNV Amplification SNV Deletion 19

44 SWI/SNF complex a SWI / SNF Complex Coverage: 168% (523 / 3110 samples) = 5 samples ARID1A (194) PBRM1 (169) SMARCA4 (67) ARID2 (56) *ARID1B (41) **ADNP (21) SMARCB1 (19) SMARCA4 LUAD: 8x10-3 ARID1A UCEC: 158x10-15 BLCA: 488x10-9 PBRM1 KIRC: 7x10-97 S10T N80Y E226Q K231R I245F I252R P427S N453S F456S R522L M523R N528I Y580C E583K M586K M586I M586T N601K L614V L618H G626V I709T Y718C M731K R760H E860K R876L P1048L P1048R R1088Q V1136L T1202K G1231R C1233W E1248G E1287D I1309T D1325N R1411C W1417G M1539I G1554R G1577C L1611P R1670P ARID1B ADNP ARID2 SMARCB1 BLCA: 001 All PPI networks MultiNet irefindex HINT HI missense Scale (AA) PBRM1 BAH_dom Bromodomain HMG_superfamily Legend Legend Cancer types BLCA BRCA COADREAD GBM HNSC KIRC LAML LUAD LUSC OV UCEC Mutation types Inactivating SNV Amplification SNV Deletion 19

45 SWI/SNF complex a SWI / SNF Complex Coverage: 168% (523 / 3110 samples) = 5 samples ARID1A (194) PBRM1 (169) SMARCA4 (67) ARID2 (56) *ARID1B (41) **ADNP (21) SMARCB1 (19) SMARCA4 LUAD: 8x10-3 ARID1A UCEC: 158x10-15 BLCA: 488x10-9 PBRM1 KIRC: 7x10-97 S10T N80Y E226Q K231R I245F I252R P427S N453S F456S R522L M523R N528I Y580C E583K M586K M586I M586T N601K L614V L618H G626V I709T Y718C M731K R760H E860K R876L P1048L P1048R R1088Q V1136L T1202K G1231R C1233W E1248G E1287D I1309T D1325N R1411C W1417G M1539I G1554R G1577C L1611P R1670P ARID1B ADNP ARID2 SMARCB1 BLCA: 001 All PPI networks MultiNet irefindex HINT HI missense Scale (AA) PBRM1 BAH_dom Bromodomain HMG_superfamily Wilson and Roberts Nature Reviews Cancer (2011) Involved in nucleosome remodeling ACTL6A/B SMARCD 1/2/3 SMARCC1 PBAF SMARCA4 SMARCB1 SMARCC2 ARID2 PBRM1 DPF 1/2/3 SMARCE1 BRD7 ACTL6A/B SMARCD 1/2/3 SMARCC1 BAF SMARCA 2/4 SMARCB1 SMARCC2 ARID1A/B DPF 1/2/3 SMARCE1 Legend Legend Cancer types BLCA BRCA COADREAD GBM HNSC KIRC LAML LUAD LUSC OV UCEC Mutation types Inactivating SNV Amplification SNV Deletion 19

46 Cohesin and condensin complexes a Cohesin Complex Coverage: 74% (229 / 3110 samples) = 5 samples STAG2 (59) SMC1A (44) **PDS5B (33) **STAG1 (31) **PDS5A (26) *RAD21 (23) **WAPAL (23) STAG2 BLCA: 5x10-3 Cohesin complex 4/5 members of complex Involved in sister chromatid cohesion and gene regulation Mutated in >4% of samples in each cancer type STAG1 SMC1A H243N S290C A428V L525R K581E Q619R S637fs V666L A704P D706G Q815* N941I R958H R958C Q994K E995Q L1001fs L1001Q exon27+2 E1027K D1059H D1059A V1075L Q1116* RAD21 PDS5B WAPAL PDS5A All PPI networks MultiNet irefindex HINT+HI frame_shift_del missense nonsense splice_site Scale (AA) STAG1 STAG Legend Legend Cancer types BLCA BRCA COADREAD GBM HNSC KIRC LAML LUAD LUSC OV UCEC Mutation types Inactivating SNV Amplification SNV Deletion 20

Cohesin and condensin complexes a Cohesin Complex Coverage: 74% (229 / 3110 samples) = 5 samples STAG2 (59) SMC1A (44) **PDS5B (33) **STAG1 (31) **PDS5A (26) *RAD21 (23) **WAPAL (23) STAG2 BLCA:

47 Cohesin and condensin complexes a Cohesin Complex Coverage: 74% (229 / 3110 samples) = 5 samples STAG2 (59) SMC1A (44) **PDS5B (33) **STAG1 (31) **PDS5A (26) *RAD21 (23) **WAPAL (23) STAG2 BLCA: 5x10-3 Cohesin complex 4/5 members of complex Involved in sister chromatid cohesion and gene regulation Mutated in >4% of samples in each cancer type STAG1 SMC1A H243N S290C A428V L525R K581E Q619R S637fs V666L A704P D706G Q815* N941I R958H R958C Q994K E995Q L1001fs L1001Q exon27+2 E1027K D1059H D1059A V1075L Q1116* RAD21 PDS5B WAPAL PDS5A All PPI networks MultiNet irefindex HINT+HI frame_shift_del missense nonsense splice_site Scale (AA) STAG1 STAG b Condensin Complex Coverage: 42% (131 / 3110 samples) = 5 samples **NCAPD3 (35) *SMC4 (27) **NCAPG2 (23) **NCAPH2 (19) **SMC2 (17) **NCAPD2 (14) NCAPG2 LUSC: 0015 SMC4 NCAPD3 exon2-2 I145V R176W A205V T217S P252S A369V R378Q K398T Q408* E414* R453S P529L R551S R551P S556F G569E Condensin complex 6/8 members of complex Involved in sister chromatid condensation and gene regulation Somatic mutations and expression validated using whole-genome sequencing and RNA-Seq Legend Legend Cancer types NCAPH2 SMC2 NCAPD2 All PPI networks MultiNet irefindex HINT+HI2012 BLCA BRCA COADREAD GBM HNSC KIRC LAML LUAD LUSC OV UCEC missense nonsense splice_site Mutation types Scale (AA) NCAPH2 Condensin_II_H2-like Inactivating SNV Amplification SNV Deletion 20

48 Outline 1 A new algorithm, HotNet2 2 Application to TCGA Pan-Cancer data A101D A101* A101E L106T P179V E187G R252L V276T 3 Comparison of HotNet2 to similar methods 21

49 HotNet2 outperforms other methods on real data No gold standard dataset compare methods at identifying putative cancer genes Dataset of putative cancer genes Cancer genes have: 1 20% truncating mutations; or, 2 20% mutations clustered at a locus (c) Vogelstein et al (Science, 2013) TPR = Sensitivity FPR = 1-Specificity 22

50 Summary HotNet2: Novel algorithm that analyzes topology and mutations simultaneously with asymmetric heat diffusion Identifies known and novel pathways and complexes with frequently and rarely mutated genes on TCGA Pan-Cancer data Future work: Alternate graph partitioning algorithms? Other applications: gene expression, GWAS, social networks, etc 23

51 Summary HotNet2: Novel algorithm that analyzes topology and mutations simultaneously with asymmetric heat diffusion v u Identifies known and novel pathways and complexes with frequently and rarely mutated genes on TCGA Pan-Cancer data Future work: Alternate graph partitioning algorithms? Other applications: gene expression, GWAS, social networks, etc 23

Acknowledgements Research Group Ben Raphael Fabio Vandin

Alexandra Papoutsaki Jacob Thomas Younhun Kim Collaborators

Getz Nuria Lopez-Bigas Abel Gonzalez-Perez David Tamborero

52 Acknowledgements Research Group Ben Raphael Fabio Vandin Hsin-Ta Wu Jason R Dobson Matt Reyna Jonathan Eldridge Alexandra Papoutsaki Jacob Thomas Younhun Kim Collaborators Beifang Niu Michael McLellan Li Ding Michael Lawrence Gad Getz Nuria Lopez-Bigas Abel Gonzalez-Perez David Tamborero Yuwei Chang Greg Ryslik Funding & Data NSF Travel Fellowship to RECOMB 2015

Network-based interpreta1on of perturba1ons to living systems

Network-based interpreta1on of perturba1ons to living systems Sushmita Roy sroy@biostat.wisc.edu Computa1onal Network Biology Biosta2s2cs & Medical Informa2cs 826 Computer Sciences 838 hbps://compnetbiocourse.discovery.wisc.edu