Integrating Multi-Omics Data to Identify Cancer Driver Modules Supplementary Information

ModulOmics: Integrating Multi-Omics Data to Identify Cancer Driver Modules Supplementary Information

Dana Silverbush∗†1, Simona Cristea*†2,3,4, Gali Yanovich5, Tamar Geiger5, Niko Beerenwinkel‡6,7, and Roded Sharan‡1

1Blavatnik School of Computer Science, Tel Aviv University, Israel 2Department of Biostatistics and Computational Biology, Dana-Farber Cancer Institute, Boston, Massachusetts, USA 3Department of Biostatistics, Harvard T.H. Chan School of Public Health, Boston, Massachusetts, USA 4Department of Stem Cell and Regenerative Biology, Harvard University, Cambridge, Massachusetts, USA 5Department of Human Molecular Genetics and Biochemistry, Sackler Faculty of Medicine, Tel Aviv University, Israel 6Department of Biosystems Science and Engineering, ETH Zurich, Basel, Switzerland 7Swiss Institute of Bioinformatics, Basel, Switzerland

Contents

1 Data processing 2 1.1 Data acquisition ...... 2 1.2 Criteria for receptor-based classiﬁcation ...... 2

2 Comparison with other tools 3 2.1 HotNet2 ...... 3 2.2 TiMEx ...... 3 2.3 TieDIE ...... 3 2.4 MEMCover ...... 3

3 Extended Results 4 3.1 Module size statistics ...... 4 3.2 Sensitivity analyses ...... 5 3.3 Driver modules are enriched with cancer drivers ...... 8 3.3.1 Positive and negative control lists ...... 8 3.3.2 Driver enrichment in top modules ...... 8 3.3.3 Driver genes enrichment in top unique genes ...... 8 3.4 Driver modules are functionally coherent ...... 15 3.4.1 Pathways per module size ...... 15 3.4.2 Pathway enrichment in subsets of three omics ...... 15 3.4.3 Module members participating in enriched pathways ...... 17 3.5 Extended analysis of breast cancer subtypes ...... 18 3.6 Demonstrating the generality of ModulOmics via complexes detection ...... 27

∗equal contribution †corresponding author ‡equal contribution

1 1 Data processing 1.1 Data acquisition ModulOmics identiﬁes driver modules on the basis on DNA and RNA cancer patient data, integrated with PPI networks and known regulatory connections. We analyzed patient data from the TCGA project [2, 3, 4], downloaded from cBio portal [5]. We integrated the patient data with the PPI network Hippie [18] containing 238, 165 physical interactions, and with the regulatory connections from the database Transcriptional Regulatory Relationships Unraveled by Sentence-based Text Mining (TRRUST) [9] containing 8, 908 Tran- scription Factor (TF)-target regulatory relationships of 821 human TFs. To further evaluate functional connections in the triple negative (TN) breast cancer subtype, we used the reverse phase protein array (RPPA) data published by the TCGA. In addition, we evaluated the power of the highest ranking modules in distinguishing healthy tissues from cancerous ones using an independent mass-spectrometry dataset containing over 62 samples of Luminal A and healthy tissue [16, 20]. Proteins were quantiﬁed using the method Super-SILAC [8].

1.2 Criteria for receptor-based classification Subtyping of breast cancer in clinical practice is mostly done by immunohistochemistry (IHC), based on the expression of estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (Her2). In this study, we classified the samples based on the information available in TCGA. The ER positive group was stratified based on Her2 expression, into Luminal B subtype for cases with positive receptor expression, and Luminal A for cases with negative expression. The ER negative group was segregated based on Her2 expression as Her2-amplified subtype in case of positive expression and TN subtype in case of negative expression.

2 2 Comparison with other tools 2.1 HotNet2 HotNet2 [13] identifies subnetworks of a PPI that contain genes with significant numbers of mutations. HotNet2 uses a localized heat diffusion process to combine the mutational genomic data with the PPI data. To calculate statistical significance, HotNet2 uses a permutation test in which the gene scores are permuted among the genes, followed by a two-stage statistical test: first, a p-value for the number of subnetworks in the list is computed, and second, the estimate of the false discovery rate of the list of subnetworks is obtained. As recommended by the authors, we used SNVs and CNAs as the prior set, and assigned the initial prior score of each genetic alteration to be its alteration frequency in the data. We applied HotNet2 on the same PPI network we used with ModulOmics. To assign a p-value, we used 100 permuted networks as background. To calculate a hyper-geometric score for pathway enrichment with Expander [21], we used modules of up to size 7, since larger modules are more likely to be unspecific from a mechanistic perspective.

2.2 TiMEx We ran TiMEx [6] with default parameters on the same binary dataset used as input for ModulOmics, consisting of binary SNVs and CNAs alterations. We considered as significant all resulting mutually exclusive groups with Bonferroni-corrected p-value < 0.1. Even though TiMEx and the mutual exclusivity score of ModulOmics are based on the same probabilistic model, the search strategy is different for the two methods. Therefore, TiMEx and the simplified omic approach of mutual exclusivity are expected to identify different modules in the data.

2.3 TieDIE We compared our results with TieDIE [14], a method to detect one single subnetwork by incorporating mutation frequency derived from DNA data, differential expression derived from RNA data, regulatory network information, and a PPI network. TieDIE amplifies the signal from different inputs throughout the PPI network using network propagation and extracts the subnetwork casted by the amplified signals. ModulOmics differs from TieDIE in its goal, as TieDIE detects a single subnetwork, whereas ModulOmics identifies multiple modules. We ran TieDIE on the three cancer cohorts used in this study: GBM, breast and ovarian cancers, using the same PPI and regulatory networks as for ModulOmics. ModulOmics modules were consistently enriched with over 80% of driver genes and under 5% of non-driver genes, while the subnetwork inferred by TieDIE, consisting of more than 300 genes, contained less than 30% known driver genes and more than 10% known non-driver genes. In addition, the KEGG pathway enrichment scores of the modules identified by ModulOmics ranged from 160 to 231, while TieDIE scores ranged from 5 to 76, across the three cancer types.

2.4 MEMCover We ran MEMCover [12] on each of the three cancer types, with default parameters. We used the same PPI as for ModulOmics, with an edge weight threshold of 0.4. The resulting modules were separated by size and further ranked by their average coverage in each cohort.

3 3 Extended Results 3.1 Module size statistics ModulOmics identiﬁes modules of ﬁxed sizes (in this application, we used 2, 3, and 4), which are further merged and ranked according to their scores. Figure S1 shows the distribution of module sizes across the top 50 modules, while tables S1- S3 show the average single-omic and multi-omics scores for each module size.

Breast GBM Ovarian 35 25 30

30 25 20 25 20 20 15 15 Count Count Count 15 10 10 10 5 5 5

0 0 0 2 3 4 2 3 4 2 3 4 Module size Module size Module size

Figure S1: Distribution of module sizes as computed across the top 50 modules identiﬁed in breast cancer, GBM and ovarian cancer.

Table S1: Mean scores for the top 50 modules in breast cancer, by module size.

Module size ME PPI co-regulation co-expression ModulOmics 2 0.92 0.71 0.40 0.55 0.64 3 0.87 0.82 0.64 0.30 0.66 4 0.77 0.95 0.49 0.23 0.61

Table S2: Mean scores for the top 50 modules in GBM, by module size.

Module size ME PPI co-regulation co-expression ModulOmics 2 0.51 0.90 0.80 0.38 0.65 3 0.87 0.92 0.71 0.33 0.71 4 0.82 0.99 0.99 0.18 0.74

Table S3: Mean scores for top 50 modules in ovarian cancer, by module size.

Module size ME PPI co-regulation co-expression ModulOmics 2 1.00 0.85 1.00 0.26 0.78 3 0.97 0.94 1.00 0.17 0.77 4 0.84 0.99 1.00 0.21 0.76

4 3.2 Sensitivity analyses We evaluated the sensitivity of the inferred modules to diﬀerent parameter choices by running ModulOmics with changed parameter values and counting repetitions of results in the top 10 modules of size 4. We changed the parameters of the stochastic search as follows: 300 initial module seeds instead of the default 200, 15 clusters instead of the default 10, and 7 top results reported by each cluster instead of the default 5. We evaluated the following metrics: i) the repetition of gene connections, i.e. gene pairs co-residing in the same module, and ii) the repetition of the gene pool reported by the top modules, regardless of which module they belonged to. Across the three cancer types, the majority of gene pairs remained connected: 24, 19 and 16 gene pairs in breast cancer, GBM, and ovarian cancer, respectively, were common to the top modules inferred with each parameter conﬁguration. Additionally, the majority of genes collectively included in the top modules were robust to parameter changes: 21, 8 and 8 genes respectively (Figure S2).

A Gene pairs in top modules

Breast GBM Ovarian

7 top results 15 clusters 7 top results 15 clusters 7 top results 15 clusters

3 1 10 9 9 5 5 12 6 24 19 16 2 6 4 2 0 0

9 4 14

300 initial seeds 300 initial seeds 300 initial seeds

B Genes in top modules

Breast GBM Ovarian

7 top results 15 clusters 7 top results 15 clusters 7 top results 15 clusters 1 1 1 3 1 3 3 4 1 11 8 8 1 2 2 1 0 0 3 1 3 300 initial seeds 300 initial seeds 300 initial seeds

Figure S2: Robustness of top 10 modules inferred by ModulOmics. A) Intersection of gene pairs co-residing in the same module, for various parameter choices. B) Intersection of the gene pool collectively reported by the top 10 modules, for various parameter choices.

We further calculated how many suggested gene exchanges were accepted in the stochastic search (Ta- ble S4). As expected, for modules of size 2, no suggested exchanges were accepted, since the ILP solution recovers the pairwise global maximum. For modules of sizes 3 or 4, an average of 0.69 exchanges per module were accepted. The inferred modules were highly robust to increasing the number of suggested exchanges, such that using 30 exchanges yielded highly similar modules (data not shown).

5 Table S4: Gene exchanges made in the stochastic search, while reﬁning the modules. The average # of accepted exchanges shown is per module.

# exchanges Cancer Module size Accepted exchanges Rejected exchanges Average # of accepted exchanges 20 exchanges Breast 3 11 869 0.35 20 exchanges Breast 4 23 917 0.7 20 exchanges GBM 3 17 863 0.35 20 exchanges GBM 4 34 866 0.72 20 exchanges Ovarian 3 71 1209 1.39 20 exchanges Ovarian 4 93 1307 1.82

We next evaluated the contribution of the stochastic search to the functional connectivity of the resulted modules by exploring the properties of their enrichment with known pathways (see details about pathway enrichment calculation in section 3.4). To this end, we compared the mean number of module members in an enriched pathway, as well as the mean enrichment score, before and after the stochastic search (Figure S3). We concentrated on the top 30 modules of largest size (size 4, with the highest rate of accepted exchanges during the stochastic search, as demonstrated by Table S4). Reassuringly, both metrics showed that the stochastic search improves the identiﬁcation of functionally connected groups. Table S5 shows the ﬁxed parameters used by ModulOmics in computing the individual scores, as well as in the stochastic search.

Table S5: Fixed parameters used by ModulOmics.

Context Parameter Default value Explanation Co-Expression test Expression quantile k = 2 A gene is deﬁned as expressed if its q = 4 expression averaged across all samples is above the kth q-quantile. Genes that are unexpressed are also not co-expressed.

Co-regulation test active TF |zScore| > 1 Active TF in the cohort in 25% of the samples

Module search Initial seeds 300 Number of initial modules inferred by the ILP based on pairwise scores. These seeds are then further reﬁned by the stochastic search to maximize the global ModulOmics score.

Module search Number of clusters 10 Number of clusters in which the seed modules are divided, so as to start the search each time from a diﬀerent cluster.

Module search Top results 5 Top modules reported by each cluster. The top modules are aggregated to one list which is re-ranked and retrieved to the user.

In this section, we showed that the modules identiﬁed by ModulOmics are highly robust to changes in the module search parameters. To choose default parameter values for the co-regulation and co-expression tests

6 A

2.8

2.6 2.6 2.5 2.5 2.4 2.4

2.2 2.1

2.0 an enriched pathway an enriched 2.0 Mean # of Mean members module

1.8 Breast GBM Ovarian

190

180 178.4

170 166.0 162.7 163.0 160 159.6

151.1 150 Mean enrichment factor enrichment Mean

140 Breast GBM Ovarian

000 ModulOmics ModulOmics without stochastic search

Figure S3: Pathway enrichment analysis for ModulOmics vs. ModulOmics without using the stochastic search for the top 30 modules of size 4. A) Mean number of genes in a module to participate in a signiﬁcantly enriched pathway. B) Mean enrichment factor score per module. we scanned a range of possible values, and chose values which included a substantial amount of information for all the studied cohorts. For example, the default values chosen for the co-regulation test yielded for GBM, breast and ovarian cancer 204, 195 and 232 active transcription factors respectively. Increasing the zScore threshold to 2 (as traditionally used by TCGA) yielded only 1, 2 and 9 respectively, substantially reducing the information retrieved. Similarly, the default parameter value for the co-expression test, namely calling a gene as expressed in the cohort if its expression averaged across all samples was above the 2nd 4th-quantile, yielded for GBM, breast and ovarian cancer 14, 823, 9, 264 and 4, 064 comparable pairs respectively (out of 263, 682, 79, 401 and 33, 153 potential gene pairs). Making the threshold more stringent by setting it to the 3rd 4th-quantile yielded only 3, 690,2, 328 and 1, 008 comparable pairs. Having explored the values for the three diﬀerent cohorts, we recommend the default values described in Table S5, yet encourage the user to test a range of values when running ModulOmics on a new cohort.

7 3.3 Driver modules are enriched with cancer drivers 3.3.1 Positive and negative control lists To evaluate the enrichment of modules with known driver genes, we used a large number of control lists, derived from independent sources. As positive controls, we used the following lists, introduced in [10]: i) The Cancer Gene Census (CGC) version 73 (PosSomatic and PosTrans), a set of 569 genes manually curated by The Sanger Institute, which have alterations in somatic and germline SNVs, CNVs and translocations; ii) UniprotKB [22] (PosUniprotKB), a manually curated database of 412 functional proteins, classified as proto- oncogene, oncogene and tumour suppressor gene; iii) a query of DISEASES [15] (PosTextMine), a database of disease-gene associations extracted mainly from text-mining, which consists of 711 genes associated with cancer; and iv) The Atlas of Genetics and Cytogenetics in Oncology and Hematology (PosAGO) [11], a list of 1, 430 cancer genes manually curated by a collaborative effort spanning multiple centers. PosUnionAll is the union of all these positive control lists. In addition, we used the Network of Cancer Genes (NCG5) [1], a manually curated list consisting of 1, 571 protein-coding cancer driver genes compiled by The Sanger Institute. The gene members of the two shortest CGC lists, germline SNVs (38 genes) and CNVs (15 genes), were not identified in any high scoring module by any of the tested methods, hence are not shown here. To evaluate the enrichment of modules with known non-driver genes, we used the following lists introduced in [10], as negative controls: i) a list derived from AGO [11] consisting of 9, 457 genes that have no evidence of association with cancer (NegAgoFull); ii) a conservative version of the negative AGO list (NegAGOClean), created by filtering genes that are part of any cancer-related pathway from the MSigDB database [19], resulting in 3, 272 genes, and iii) a list of known non-driver genes introduced in [7] (NegDavoli).

3.3.2 Driver enrichment in top modules According to the enrichment of the highest scoring driver modules with known driver genes (positive controls) and known non-driver genes (negative controls), ModulOmics generally outperformed the three competitive methods evaluated: TiMEx [6], HotNet2 [13], and MEMCover [12], as well as the four single-omics and the four three-omics approaches, for groups of sizes 2, 3 and 4 (Figures S4 and S5). MEMCover identified only 3 modules of size 3 in breast cancer, and 1 module in GBM and ovarian cancer, as well as only 1 module of size 4 in any of the three cancer types; TiMEx identified only 3 modules of size 3 in ovarian cancer, 1 module of size 4 in breast cancer, and no modules in GBM or ovarian cancer; HotNet2 identified only 10 modules of size 2 in breast cancer, 8 modules in GBM, and 2 modules in ovarian cancer, only 5 modules of size 3 in breast cancer, 2 modules in GBM, and no modules in ovarian, as well as no modules of size 4 in any of the three cancer types. These performances per module size are therefore not shown in Figure S4. Table S6 displays the scores plotted in Figures 2, S4 and S5B.

3.3.3 Driver genes enrichment in top unique genes One of the features of ModulOmics is that each gene can participate in multiple modules (Figure S6). ModulOmics outperformed the other methods also when restricting the comparison to the same number of unique genes part of the top modules, regardless of their membership (Figure S9).

8 A) Breast size 2 size 3 size 4 ● 1.0 1.0 1.0 ● ● ● ● ● ● 0.8 0.8 0.8

● ● 0.6 0.6 0.6 ●

e driver list ● ● e driver list ● e driver list 0.4 0.4 0.4 ● ● ● 0.2 0.2 0.2 positi v positi v positi v 0.0 0.0 0.0

0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 negative driver list negative driver list negative driver list

B) GBM size 2 size 3 size 4 ● ● ● 1.0 1.0 1.0 ● ● 0.8 0.8 0.8 ● ● ● ● 0.6 0.6 ● 0.6 ● ● ● e driver list e driver list e driver list 0.4 0.4 0.4 ● 0.2 0.2 0.2

positi v ● positi v positi v ● 0.0 0.0 0.0 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 negative driver list negative driver list negative driver list

C) Ovarian size 2 size 3 size 4

● ● ● 1.0 1.0 1.0 ● ● ● ● 0.8 0.8 ● 0.8 ● 0.6 0.6 0.6

e driver list ● ● e driver list ● e driver list ● 0.4 0.4 0.4 ● ● ● 0.2 0.2 0.2 positi v positi v positi v 0.0 0.0 0.0

0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 negative driver list negative driver list negative driver list

ModulOmics PPI ME CoReg CoExp TiMEx HotNet2 MEMCover top 5 groups top 10 groups top 15 groups top 20 groups top 30 groups

Figure S4: The driver modules inferred by ModulOmics are enriched with cancer driver genes (extending Figure 2B in the main text). Detailed driver and non-driver enrichment scores for the positive driver list PosUnionAll and the negative driver list NegAGOClean for the eight methods assessed for the top scoring 5, 10, 15, 20 and 30 modules and for the cancer types A) breast, B) GBM and C) ovarian.

9 A Breast GBM Ovarian

NCG5 PosAGO PosUnionAll PosSomatic PosUniprotKB PosTextMine

PosTrans

NegAgoClean NegAgoFull NegEllidge — ME — ME — ME — PPI — PPI — PPI — CoExp — CoExp — CoExp — CoReg — CoReg — CoReg ModulOmics ModulOmics ModulOmics

0 0.2 0.4 0.6 0.8 1 0 0.05 0.1 0.15 0.2 percentage in positive lists percentage in negative lists

Breast GBM Ovarian B ● ●

● 1.0 1.0 ● 1.0 ● ● ● ● ● 0.8 0.8 0.8 ● ● e list e list e list

● 0.6 0.6 0.6 0.4 positi v positi v 0.4 positi v 0.4 0.2 0.2 0.2 0.0 0.0 0.0 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 negative list negative list negative list

ModulOmics —CoExp —CoReg —ME —PPI top 5 groups top 10 groups top 15 groups top 20 groups top 30 groups

Figure S5: The driver modules inferred by ModulOmics are enriched with cancer driver genes (extending Figure 2 in the main text). A) The average driver enrichment (red heatmaps) and non-driver enrichment (blue heatmaps) across the top 10 scoring modules inferred by ModulOmics and by using only subsets of three omics data sources. B) Detailed driver and non-driver enrichment scores for the positive driver list PosUnionAll and the negative driver list NegAGOClean for ModulOmics and the subsets of three omics for the top scoring 5, 10, 15, 20 and 30 modules. The − sign next to an omic data source represents the situation in which that omic data source was excluded from the computation of the ModulOmics score.

10 Breast GBM Ovarian 24 ERBB3 PTEN AKT1 LAMA2 25 ASAP1 WDFY3 BRCA2 MAP3K1 MYC NF1 MTA1 CDK1 DNAH7 32 20 PPEF1 SETD2 CDK2 ATM RB1 20 MAP2 GOPC CDKN1A GATA3 DLC1 CDK4 16 CREBBP BRCA2 CCNE1 ATN1 24 BCL2L1 DMD PDGFRA 15 CDH1 EGFR TP53 AKT1 CDK6 FOXA1 gene gene 12 gene KRAS LPA CDKN2A CACNA1F HDAC2 RB1 NID1 16 F5 TP53 10 BRCA1 ROR1 CD33 CTCF 8 CCNG1 CCNA2 DCC CUL7 MYB CCND1 MYLK PTPRK 8 TP53 RPL5 IGF1R 5 ERBB2 4 MADD PIK3CA MYC RUNX1 PARP1 ARID1A CDH1 MYB ARHGEF7 count count count

Figure S6: Occurrence of genes in the top 50 modules identiﬁed by ModulOmics across the three cancer types.

Breast GBM Ovarian 0.75 1.0 0.75 NCG5 NCG5 NCG5

PosAGO 0.60 PosAGO 0.8 PosAGO 0.60

PosUnionAll PosUnionAll PosUnionAll 0.45 0.6 0.45 PosSomatic PosSomatic PosSomatic 0.30 0.4 PosUniprotKB PosUniprotKB PosUniprotKB 0.30

PosTextMine 0.15 PosTextMine 0.2 PosTextMine 0.15 PosTrans PosTrans PosTrans 0.00 0.0 ModulOmics MemCover HotNet2 TiMEx ModulOmics MemCover HotNet2 TiMEx ModulOmics MemCover HotNet2 TiMEx

Breast GBM Ovarian 0.6 0.5 0.40 NegAgoClean 0.5 NegAgoClean NegAgoClean 0.4 0.32 0.4 0.3 0.24 NegDavoli 0.3 NegDavoli NegDavoli 0.2 0.2 0.16

0.1 NegAgoFull 0.1 NegAgoFull NegAgoFull 0.08

0.0 0.0 0.00 ModulOmics MemCover HotNet2 TiMEx ModulOmics MemCover HotNet2 TiMEx ModulOmics MemCover HotNet2 TiMEx

Figure S7: The 20 unique genes in the top driver modules inferred by ModulOmics are more enriched with cancer driver genes.

11 Table S6: Scores corresponding to the performance of the methods displayed in Figures 2B, S4 and S5B. Each pair (a, b) represents the average driver enrichment scores for the positive control list PosUnionAll and the negative control list NegAGOClean.

Figure Size Method Top Breast cancer GBM Ovarian cancer 2B ALL ModulOmics 5 (0.95, 0) (1, 0) (1, 0) 2B ALL ModulOmics 10 (0.83, 0) (1, 0) (1, 0) 2B ALL ModulOmics 15 (0.82, 0) (1, 0) (0.96, 0) 2B ALL ModulOmics 20 (0.80, 0) (1, 0) (0.95, 0) 2B ALL ModulOmics 30 (0.83, 0) (1, 0) (0.95, 0) S4 2 ModulOmics 5 (1, 0) (1, 0) (1, 0) S4 2 ModulOmics 10 (0.85, 0) (1, 0) (1, 0) S4 2 ModulOmics 15 (0.9, 0) (1, 0) (0.96, 0) S4 2 ModulOmics 20 (0.9, 0) (1, 0) (0.97, 0) S4 2 ModulOmics 30 (0.85, 0) (0.95, 0) (0.96, 0) S4 3 ModulOmics 5 (0.8, 0) (1, 0) (0.93, 0) S4 3 ModulOmics 10 (0.73, 0) (1, 0) (0.96, 0) S4 3 ModulOmics 15 (0.8, 0) (1, 0) (0.95, 0) S4 3 ModulOmics 20 (0.83, 0) (0.98, 0) (0.93, 0) S4 3 ModulOmics 30 (0.82, 0) (0.97, 0) (0.93, 0) S4 4 ModulOmics 5 (0.85, 0) (1, 0) (1, 0) S4 4 ModulOmics 10 (0.8, 0) (1, 0) (0.95, 0) S4 4 ModulOmics 15 (0.85, 0) (1, 0) (0.93, 0) S4 4 ModulOmics 20 (0.88, 0) (1, 0) (0.93, 0) S4 4 ModulOmics 30 (0.86, 0) (1, 0) (0.94, 0) 2B ALL PPI 5 (0.8, 0) (0.93, 0) (0.78, 0) 2B ALL PPI 10 (0.85, 0) (0.84, 0) (0.79, 0) 2B ALL PPI 15 (0.87, 0) (0.79, 0) (0.79, 0) 2B ALL PPI 20 (0.87, 0) (0.8, 0) (0.79, 0) 2B ALL PPI 30 (0.83, 0) (0.81, 0) (0.79, 0) S4 2 PPI 5 (0.9, 0) (0.9, 0) (0.8, 0) S4 2 PPI 10 (0.85, 0) (0.9, 0) (0.75, 0) S4 2 PPI 15 (0.76, 0) (0.86, 0) (0.8, 0) S4 2 PPI 20 (0.82, 0) (0.8, 0) (0.8, 0) S4 2 PPI 30 (0.81, 0) (0.7, 0) (0.85, 0) S4 3 PPI 5 (0.8, 0) (0.86, 0) (0.8, 0) S4 3 PPI 10 (0.86, 0) (0.76, 0) (0.8, 0) S4 3 PPI 15 (0.84, 0) (0.75, 0) (0.84, 0) S4 3 PPI 20 (0.83, 0) (0.76, 0) (0.81, 0) S4 3 PPI 30 (0.83, 0) (0.71, 0) (0.85, 0) S4 4 PPI 5 (0.95, 0) (0.75, 0) (0.8, 0) S4 4 PPI 10 (0.87, 0) (0.75, 0) (0.82, 0) S4 4 PPI 15 (0.88, 0) (0.73, 0) (0.85, 0) S4 4 PPI 20 (0.88, 0) (0.7, 0) (0.83, 0) S4 4 PPI 30 (0.85, 0) (0.66, 0) (0.85, 0) 2B ALL ME 5 (0.4, 0.06) (0.2, 0) (0.4, 0.06) 2B ALL ME 10 (0.4, 0.13) (0.25, 0) (0.33, 0.1) 2B ALL ME 15 (0.4, 0.08) (0.3, 0) (0.35, 0.08) 2B ALL ME 20 (0.36, 0.1) (0.3, 0) (0.33, 0.08) 2B ALL ME 30 (0.35, 0.1) (0.3, 0) (0.31, 0.08) S4 2 ME 5 (0.4, 0.1) (0.2, 0) (0.3, 0.1) S4 2 ME 10 (0.45, 0.1) (0.25, 0) (0.4, 0.1) S4 2 ME 15 (0.46, 0.13) (0.3, 0) (0.43, 0.13) S4 2 ME 20 (0.5, 0.12) (0.3, 0) (0.45, 0.12

12 S4 2 ME 30 (0.5, 0.08) (0.3, 0) (0.38, 0.16) S4 3 ME 5 (0.4, 0.06) (0.46, 0) (0.4, 0.06) S4 3 ME 10 (0.4, 0.13) (0.3, 0) (0.33, 0.1) S4 3 ME 15 (0.4, 0.08) (0.28, 0.02) (0.35, 0.08) S4 3 ME 20 (0.36, 0.1) (0.3, 0.01) (0.33, 0.08 S4 3 ME 30 (0.35, 0.1) (0.31, 0.01) (0.31, 0.08) S4 4 ME 5 (0.25, 0.1) (0.45, 0.15) (0.4, 0.15) S4 4 ME 10 (0.25, 0.1) (0.37, 0.1) (0.4, 0.2) S4 4 ME 15 (0.23, 0.15) (0.41, 0.1) (0.36, 0.23) S4 4 ME 20 (0.25, 0.16) (0.38, 0.07) (0.42, 0.18 S4 4 ME 30 (0.24, 0.15) (0.37, 0.05) (0.39, 0.19) 2B ALL CoReg 5 (0.93, 0) (1, 0) (0.7, 0) 2B ALL CoReg 10 (0.93, 0) (0.95, 0) (0.8, 0) 2B ALL CoReg 15 (0.93, 0) (0.96, 0) (0.76, 0) 2B ALL CoReg 20 (0.95, 0) (0.97, 0) (0.82, 0) 2B ALL CoReg 30 (0.95, 0) (0.98, 0) (0.8, 0) S4 2 CoReg 5 (1, 0) (1, 0) (0.7, 0) S4 2 CoReg 10 (0.95, 0) (0.95, 0) (0.8, 0) S4 2 CoReg 15 (0.93, 0) (0.96, 0) (0.76, 0) S4 2 CoReg 20 (0.92, 0) (0.97, 0) (0.82, 0) S4 2 CoReg 30 (0.94, 0) (0.98, 0) (0.8, 0) S4 3 CoReg 5 (0.93, 0) (1, 0) (1, 0) S4 3 CoReg 10 (0.93, 0) (1, 0) (0.96, 0) S4 3 CoReg 15 (0.93, 0) (1, 0) (0.97, 0) S4 3 CoReg 20 (0.91, 0) (1, 0) (0.95, 0) S4 3 CoReg 30 (0.9, 0) (1, 0) (0.94, 0) S4 4 CoReg 5 (0.95, 0) (1, 0) (1, 0) S4 4 CoReg 10 (0.9, 0.025) (1, 0) (0.97, 0) S4 4 CoReg 15 (0.91, 0.01) (1, 0) (0.96, 0) S4 4 CoReg 20 (0.92, 0.01) (1, 0) (0.97, 0) S4 4 CoReg 30 (0.93, 0.08) (1, 0) (0.97, 0) 2B ALL CoExp 5 (0.38, 0) (0.5, 0.16) (0.38, 0) 2B ALL CoExp 10 (0.32, 0.05) (0.51, 0.1) (0.32, 0.05) 2B ALL CoExp 15 (0.23, 0.06) (0.55, 0.07) (0.28, 0.05) 2B ALL CoExp 20 (0.21, 0.07) (0.47, 0.1) (0.24, 0.07) 2B ALL CoExp 30 (0.19, 0.12) (0.43, 0.1) (0.23, 0.1) S4 2 CoExp 5 (0.4, 0) (0.6, 0.1) (0.4, 0.1) S4 2 CoExp 10 (0.2, 0.1) (0.35, 0.15) (0.25, 0.1) S4 2 CoExp 15 (0.13, 0.16) (0.33, 0.1) (0.26, 0.13) S4 2 CoExp 20 (0.1, 0.22) (0.35, 0.075) (0.25, 0.15) S4 2 CoExp 30 (0.08, 0.23) (0.35, 0.06) (0.2, 0.2) S4 3 CoExp 5 (0.26, 0) (0.53, 0.06) (0.26, 0) S4 3 CoExp 10 (0.2, 0.06) (0.5, 0.06) (0.15, 0.13) S4 3 CoExp 15 (0.15, 0.11) (0.53, 0.08) (0.2, 0.11) S4 3 CoExp 20 (0.13, 0.11) (0.48, 0.06) (0.2, 0.11) S4 3 CoExp 30 (0.1, 0.16) (0.37, 0.07) (0.2, 0.1) S4 4 CoExp 5 (0.2, 0.1) (0.6, 0.05) (0.2, 0.1) S4 4 CoExp 10 (0.2, 0.12) (0.45, 0.1) (0.22, 0.07) S4 4 CoExp 15 (0.16, 0.11) (0.4, 0.1) (0.2, 0.06) S4 4 CoExp 20 (0.16, 0.11) (0.4, 0.1) (0.16, 0.08) S4 4 CoExp 30 (0.15, 0.12) (0.41, 0.09) (0.13, 0.1) 2B ALL MEMCover 5 (0.6, 0.1) (0.7, 0) (0.55, 0) 2B ALL MEMCover 10 (0.75, 0.05) (0.7, 0) (0.42, 0.05) 2B ALL MEMCover 15 (0.65, 0.06) (0.7, 0.03) (0.35, 0.03)

13 2B ALL MEMCover 20 (0.69, 0.07) (0.57, 0.05) (0.36, 0.05) 2B ALL MEMCover 30 (0.54, 0.06) (0.46, 0.03) (0.34, 0.04) S4 2 MEMCover 5 (0.6, 0.1) (0.7, 0) (0.4, 0) S4 2 MEMCover 10 (0.65, 0.1) (0.7, 0) (0.3, 0.05) S4 2 MEMCover 15 (0.63, 0.06) (0.63, 0.03) (0.3, 0.03) S4 2 MEMCover 20 (0.57, 0.07) (0.52, 0.05) (0.3, 0.05) S4 2 MEMCover 30 (0.48, 0.06) (0.44, 0.03) (0.29, 0.04) 2B ALL TiMEx 5 (0.6, 0) (0.5, 0.2) (0.6, 0.1) 2B ALL TiMEx 10 (0.6, 0.05) (0.6, 0.1) (0.6, 0.05) 2B ALL TiMEx 15 (0.66, 0.03) (0.4, 0.15) (0.67, 0.05) 2B ALL TiMEx 20 (0.7, 0.025) (0.31, 0.11) (0.65, 0.09) 2B ALL TiMEx 30 (0.73, 0.03) (0.35, 0.09) (0.65, 0.09) S4 2 TiMEx 5 (0.6, 0) (0.5, 0.2) (0.6, 0.1) S4 2 TiMEx 10 (0.6, 0.05) (0.6, 0.1) (0.6, 0.05) S4 2 TiMEx 15 (0.66, 0.03) (0.53, 0.06) (0.63, 0.06) S4 2 TiMEx 20 (0.7, 0.025) (0.52, 0.1) (0.63, 0.06) S4 2 TiMEx 30 (0.7, 0.03) (0.52, 0.1) (0.63, 0.06) S4 3 TiMEx 5 (0.8, 0.13) (0, 0.26) S4 3 TiMEx 10 (0.73, 0.06) (0.06, 0.13) S4 3 TiMEx 15 (0.78, 0.04) (0.08, 0.11) S4 3 TiMEx 20 (0.78, 0.04) (0.08, 0.11) S4 3 TiMEx 30 (0.78, 0.04) (0.08, 0.11) 2B ALL HotNet2 5 (0.42, 0.09) (0.71, 0) (0.53, 0.1) 2B ALL HotNet2 10 (0.42, 0.11) (0.51, 0.05) (0.44, 0.09) 2B ALL HotNet2 15 (0.47, 0.1) (0.37, 0.06) (0.44, 0.09) 2B ALL HotNet2 20 (0.4, 0.08) (0.41, 0.06) (0.44, 0.09) 2B ALL HotNet2 30 (0.36, 0.07) (0.41, 0.06) (0.44, 0.09) S4 2 HotNet2 5 (0.4, 0.1) (0.1, 0.2) S4 2 HotNet2 10 (0.25, 0.05) S4 3 HotNet2 5 (0.46, 0.13) S5B ALL -CoExp 5 (1, 0) (0.9, 0) (1, 0) S5B ALL -CoExp 10 (0.93, 0) (0.86, 0) (0.96, 0) S5B ALL -CoExp 15 (0.92, 0) (0.83, 0.01) (0.97, 0) S5B ALL -CoExp 20 (0.91, 0) (0.85, 0.01) (0.93, 0) S5B ALL -CoExp 30 (0.93, 0) (0.78, 0.01) (0.89, 0) S5B ALL -CoReg 5 (1, 0) (0.88, 0.04) (0.75, 0) S5B ALL -CoReg 10 (0.97, 0.02) (0.87, 0.02) (0.81, 0) S5B ALL -CoReg 15 (0.95, 0.03) (0.85, 0.05) (0.74, 0.01) S5B ALL -CoReg 20 (0.93, 0.02) (0.83, 0.04) (0.78, 0.01) S5B ALL -CoReg 30 (0.91, 0.03) (0.81, 0.04) (0.78, 0) S5B ALL -ME 5 (0.9, 0) (0.76, 0.04) (0.77, 0) S5B ALL -ME 10 (0.8, 0.05) (0.65, 0.02) (0.63, 0) S5B ALL -ME 15 (0.72, 0.06) (0.68, 0.05) (0.65, 0.01) S5B ALL -ME 20 (0.69, 0.06) (0.72, 0.04) (0.7, 0.01) S5B ALL -ME 30 (0.69, 0.09) (0.73, 0.04) (0.72, 0) S5B ALL -PPI 5 (0.85, 0.05) (0.6, 0.04) (0.93, 0) S5B ALL -PPI 10 (0.85, 0.02) (0.66, 0.04) (0.8, 0) S5B ALL -PPI 15 (0.82, 0.01) (0.6, 0.04) (0.78, 0.01) S5B ALL -PPI 20 (0.8, 0.04) (0.62, 0.04) (0.76, 0.01) S5B ALL -PPI 30 (0.78, 0.04) (0.65, 0.05) (0.68, 0)

14 3.4 Driver modules are functionally coherent To evaluate the enrichment of modules with known pathways, we used two statistical tests of module-pathway intersection, as proposed by the Expander software [21]: i) a hyper-geometric enrichment test to calculate the occurrence probability of the intersection of a module with a pathway at random when drawing from all gene coding proteins, and ii) an enrichment factor designed to ease the bias towards larger modules. The enrichment factor is deﬁned as the ratio between the sizes of the intersection of each module and each pathway and the intersection of that pathway and the set of all background genes (all protein coding genes), normalized |module ∩ pathway| |background genes| by the sizes of the module and background respectively: |pathway ∩ background genes| × |module|

3.4.1 Pathways per module size Analyzing the pathway enrichment per module size, Figure S8 shows that no particular size dominates the enriched pathways, with the exception of modules of size 2 in ovarian cancer, which are less enriched with known pathways.

Breast GBM Ovarian 16 A 14 12 10 8 6

# significantly 4

enriched pathways 2 0 2 3 4 2 3 4 2 3 4 Module size Module size Module size

350

B 300

250

200

150

100

Enrichment factor 50

0 2 3 4 2 3 4 2 3 4 Module size Module size Module size

Figure S8: Pathway enrichment per module size based on the 30 top modules identiﬁed by ModulOmics.

3.4.2 Pathway enrichment in subsets of three omics We further evaluated the contribution of each single omic to pathway enrichment, by running ModulOmics with subsets of three omics at a time (Figure S9). We found that using all four data sources improves the identiﬁcation of functionally coherent modules in 92% of the tested cases, with the exception of removing the CoExpression test in GBM. However, in that case, even though the resulting modules were more enriched with known KEGG pathways, they were less enriched with known driver genes (Figure S5).

15 Breast GBM Ovarian 12

10.4 10 9.4 9.5 8.8 8.6 8.7 8.78.6 8.2 8.3 8.0 8.0 8 7.8 7.8 7.4 7.0 6.8 6.8 6.6 6.5 6 5.2 4.8 # significantly

enriched pathways 4 3.8 3.5 3.6 2.9 2.7 2.8 2.8 2.2 2.2 2.0 2 1.8 1.7 1.4 1.2 1.2 1.3 0.9 0.8 0.6 0.00.0 0.1 0.0 0 top 5 top 10 top 15 top 5 top 10 top 15 top 5 top 10 top 15

250

231

207 200 199 197 191 194 184 186 176 170 170 158 160 157 152 149 150 145 146 145 141 144 137 134 124 122 121 110 111 102 100 95 Enrichment factor

76 74 68 59 50 47 48 48 42 35

17 20 11 0 0 0 0 top 5 top 10 top 15 top 5 top 10 top 15 top 5 top 10 top 15

ModulOmics -CoExp -CoReg -ME -PPI

Figure S9: Pathway enrichment using subsets of three omics. The − sign next to an omic data source represents the situation in which that omic data source was excluded from the computation of the ModulOmics score.

16 3.4.3 Module members participating in enriched pathways

Table S7: Genes in top modules participating in enriched pathways (e.p.).

# cancer top ModulOmics HotNet2 TiMEx MEMCover modules # # # # genes # total ratio genes # total ratio genes # total ratio genes # total ratio in genes in genes in genes in genes e.p. e.p. e.p. e.p. Breast 5 4 5 0.8 6 20 0.3 2 7 0.28 6 12 0.5 Breast 10 4 7 0.57 8 30 0.26 2 12 0.16 8 23 0.34 Breast 15 5 10 0.5 8 40 0.2 3 17 0.17 11 35 0.31 GBM 5 5 6 0.83 9 18 0.5 0 9 0.0 9 11 0.81 GBM 10 7 9 0.77 15 26 0.57 0 14 0.0 13 21 0.61 GBM 15 8 11 0.72 17 30 0.56 0 19 0.0 17 31 0.54 Ovarian 5 5 5 1.0 3 9 0.33 2 7 0.28 3 12 0.25 Ovarian 10 11 12 0.91 3 9 0.33 3 13 0.23 6 23 0.26 Ovarian 15 13 15 0.86 3 9 0.33 3 18 0.16 6 33 0.18

17 3.5 Extended analysis of breast cancer subtypes This section extends the Results section in the main text in which we analyzed modules inferred in breast cancer. We first applied ModulOmics on molecularly defined subtypes of breast cancer classified using the mRNA PAM50 classification. The module size distribution (across sizes 2, 3, and 4) in the top 20 modules shows that modules in Luminal B, Her2 and Basal were generally larger (4 genes), while the majority of Luminal A modules consisted of 3 genes (Figure S10A). The scores of the modules ranged from 0.65 to 0.75 (Figure S10B). The sizes of the Luminal A, Luminal B and Her2 modules were correlated to their respective scores, with smaller modules having higher score. The pair modules contained members with very high mutation frequency, such as TP53 and GATA3, and were characterized by high mutual exclusivity scores.

A B Module sizes density distribution Correlation of module size to score 0.8 Lum A Lum A Lum B Lum B 0.7 Basal Basal Her2 0.75 Her2 0.6

0.5 0.70 0.4 Density Score

0.3

0.65 0.2

0.1

0.0 0.60 2 3 4 2 3 4 Module size Module size

Figure S10: Size distribution and correlation of size to score for modules inferred in mRNA-classiﬁed breast cancer subtypes, for the top 20 modules of sizes 2, 3, and 4. A) Distribution of module sizes (curve smoothed). B) Correlation of ModulOmics scores to module size.

18 GO terms enrichment

Luminal A Luminal B Her2 Basal DNA damage response, signal transduction resulting in induction of apoptosis GO:0008630 24.22 DNA metabolic process GO:0006259 anatomical structure morphogenesis GO:0009653 apoptotic process GO:0006915 brain development GO:0007420 cell cycle GO:0007049 cell differentiation GO:0030154 cell morphogenesis GO:0000902 cell proliferation GO:0008283 chordate embryonic development GO:0043009 chromosome organization GO:0051276 cytoskeleton organization GO:0007010 double strand break repair GO:0006302 embryo development GO:0009790 hemopoiesis GO:0030097 induction of programmed cell death GO:0012502 multi organism process GO:0051704 negative regulation of apoptotic process GO:0043066 negative regulation of cellular process GO:0048523 negative regulation of gene expression GO:0010629 negative regulation of macromolecule metabolic process GO:0010605 negative regulation of nucleobase containing compound metabolic process GO:0045934 nervous system development GO:0007399 neurogenesis GO:0022008 organelle organization GO:0006996 positive regulation of apoptotic process GO:0043065 positive regulation of gene expression GO:0010628 positive regulation of metabolic process GO:0009893 positive regulation of nitrogen compound metabolic process GO:0051173 positive regulation of protein modification process GO:0031401 positive regulation of signaling GO:0023056 regulation of T cell activation GO:0050863 regulation of biological quality GO:0065008 regulation of cell cycle process GO:0010564 regulation of cell death GO:0010941 regulation of cell differentiation GO:0045595 regulation of cell proliferation GO:0042127 regulation of cellular component biogenesis GO:0044087 regulation of multicellular organismal development GO:2000026 regulation of multicellular organismal process GO:0051239 regulation of protein modification process GO:0031399 response to external stimulus GO:0009605 response to radiation GO:0009314 10.05 -ln(p-value)

Figure S11: Signiﬁcantly enriched GO pathways across the top 20 modules (extending Figure 4C in the main text), reﬂecting the aggressiveness of the Basal and Her2 subtypes, compared to Luminal A and Luminal B. Enrichment hyper-geometric p-value was computed with Expander [21]. White corresponds to absent pathways.

19 An alternative way to study breast cancer progression is by stratifying patients according to immunohistochemistry results assessing the HER2, ER and PR receptors. The module size distribution of the top 20 modules was similar to the one obtained by mRNA-based classiﬁcation (Figure S12A). Modules in Lu- minal B, Her2-enriched and TN were generally larger (size 4), while the majority of Luminal A modules consisted of 3 genes. We observed a moderate correlation of module size to its score (Figure S12B).

A B Module sizes density distribution Correlation of module size to score 1.0 0.75 Lum A Lum A Lum B Lum B TN TN 0.8 Her2-enriched 0.70 Her2-enriched

0.6 0.65 Score Density

0.4 0.60

0.2 0.55

0.0 0.50 2 3 4 2 3 4 Module size Module size

Figure S12: Size distribution and correlation of size to score for modules inferred in receptors-classiﬁed breast cancer subtypes, for the top 20 modules of sizes 2, 3, and 4. A) Distribution of module sizes (curve smoothed). B Correlation of ModulOmics scores to module size.

20 A Receptors-based classification D Occurrence frequency in top 20 modules SNV frequency CNA frequency LumA LumB Her2 TN LumA LumB Her2 TN LumA LumB Her2 TN LumA+LumB+Her2 TN AHNAK >70% ARID1A Phosphorylation activity ASPM ATM CDK1, CYCLIN B1, AKT (tumor promoters) BRCA1 1.4 BRCA2 KS p-value 1.156e-16 CDC42BPA 1.2 CDH1 CHD4 CTCF 1.0 CUL7 DCC 0.8 DSP DYNC2H1 ERBB2 0.6 ERBB3 Distribution FLNC 0.4 FOXA1 GATA3 GRIA1 0.2 GRIK2 CDK1 GRIN2B 0.0 HECW1 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 ITPR1 RPPA z-score MAP1A MTOR MUC16 MUC5B Phosphorylation activity MYB PTEN, RB1, CYCLIN D1 (tumor supressors) MYH9 NCOA3 3.0 NCOR1 KS p-value 2.95e-09 NF1 2.5 NID1 PIK3CA PIK3R1 2.0 PTEN RB1 RUNX1 1.5 SCN5A SF3B1

SPTB Distribution 1.0 SPTBN2 TAF1 TP53 0.5 TTN VCAN 0.0 WDFY3 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 ZFP36L1 <5% Frequency RPPA z-score

B C

TN KS TN Lum A TN Lum B Her2 3.0 CDK1 (TP53 target) 4.0e-10 ME PPI co-expression co-regula�on CYCLIN B1 (BRCA1 target) 2.2e-12 0 0.1 10.99 1.5 AKT PT308 (PTEN regulated) 4.0e-04 Tumor

promoters AKT PS473 (PTEN regulated) 1.5e-02 0.0 TP53 PTEN PTEN 8.4e-08 RB1 1.1e-04 TN -1.5 BRCA1 RB1 RB1 PS807-S811 8.1e-02 Tumor

supressors CYCLIN D1 (RB1-related) 3.4e-11 -3.0 Patients (TCGA) RPPA z-score Figure S13: Modules inferred in receptors-classified breast cancer subtypes highlight differences among subtypes. A) For each receptors-based subtype and for the pooled set of genes in the top 20 modules, we computed their occurrence frequency in the top 20 modules, as well as their SNV and CNA alteration frequencies across the patient cohort. These genes are enriched with known cancer drivers and pathways, and could not have been identified if relying on SNV and CNA alteration frequencies alone. White corresponds to absent genes (0% frequency). B) The TN module PTEN, BRCA1, RB1, and TP53 (ModulOmics score 0.52) shows both pairwise and group mutual exclusivity scores of 0, unique to this subtype, potentially suggestive that the accumulating activity of multiple tumor suppressors drives TN aggressiveness. C) RPPA data supports the observation that TN patients harbor lower activity of the tumor suppressors PTEN, BRCA1, RB1, and TP53, compared to other subtypes. PTEN, RB1, phosphorylated RB1 (RB1 PS807-S811) and RB1-related CYCLIN D1 show significantly lower activity, and phosphorylated AKT (AKT PS308 and AKT PS473), which is negatively regulated by PTEN, shows higher activity. TP53 and BRCA1 were not captured directly, yet their suppressed targets, CDK1 and CYCLIN B1, show lower activity. D) The tumor suppressors PTEN, BRCA1, RB1, and TP53 can be used to separate TN tissues from the other subtypes in RPPA data. CDK1, CYCLIN B1 and AKT, tumor promoters negatively regulated by TP53, BRCA1 and PTEN respectively, show significantly higher activity as a group; PTEN, RB1 and the RB1-related CYCLIN D1, tumor suppressors, show significantly lower activity as a group.

21 Table S8: Top driver modules inferred in mRNA-classiﬁed breast cancer subtypes. subtype score co-expression co-regulation ME PPI Gene1 Gene2 Gene3 Gene4 Basal 0.799 0.241 1.0 1.0 0.958 RB1 BRCA1 NF1 CREBBP Basal 0.765 0.08 1.0 1.0 0.982 RB1 BRCA1 BRCA2 PTEN Basal 0.743 0.068 1.0 1.0 0.902 ABCB1 RB1 BRCA1 BRCA2 Basal 0.741 0.263 0.75 1.0 0.95 RB1 BRCA1 BRCA2 NF1 Basal 0.741 0.08 1.0 1.0 0.884 ABCB1 BRCA1 BRCA2 PTEN Basal 0.737 0.239 0.75 1.0 0.96 FMR1 RB1 BRCA1 BRCA2 Basal 0.731 0.267 0.75 1.0 0.906 RB1 BRCA1 BRCA2 ASPM Basal 0.726 0.017 1.0 1.0 0.887 ABCB1 RB1 BRCA1 PTEN Basal 0.718 0.175 0.75 1.0 0.948 APC BRCA1 BRCA2 NF1 Basal 0.717 0.093 1.0 0.884 0.89 TP53 BRCA2 Basal 0.712 0.211 0.75 1.0 0.887 BRCA1 BRCA2 ASPM PTEN Basal 0.703 0.139 1.0 0.703 0.97 TP53 BRCA1 BRCA2 Basal 0.695 0.046 1.0 0.779 0.953 RB1 TP53 BRCA2 Basal 0.688 0.065 1.0 0.74 0.947 TP53 BRCA2 PTEN Basal 0.685 0.104 1.0 0.643 0.994 RB1 TP53 BRCA1 BRCA2 Basal 0.684 0.0 1.0 0.834 0.903 TP53 CREBBP Basal 0.679 0.0 1.0 0.772 0.945 TP53 HSPA4 CREBBP Basal 0.674 0.095 1.0 0.611 0.989 TP53 BRCA1 BRCA2 PTEN Basal 0.672 0.0 1.0 0.723 0.967 RB1 TP53 CREBBP Basal 0.67 0.0 1.0 0.694 0.988 RB1 TP53 HSPA4 CREBBP Basal 0.667 0.17 0.75 1.0 0.747 RB1 COL12A1 BRCA1 NF1 Basal 0.667 0.217 0.75 0.717 0.986 TP53 BRCA1 BRCA2 HUWE1 Basal 0.666 0.004 1.0 0.711 0.948 FMR1 TP53 HSPA4 CREBBP Basal 0.66 0.005 1.0 0.667 0.966 FMR1 RB1 TP53 CREBBP Basal 0.656 0.007 1.0 0.75 0.869 FMR1 TP53 CREBBP Basal 0.656 0.184 0.667 0.821 0.952 PRKDC TP53 BRCA2 Basal 0.653 0.108 0.75 0.778 0.975 WHSC1 TP53 BRCA2 PTEN Basal 0.649 0.123 0.75 0.747 0.978 TP53 BRCA2 HUWE1 PTEN Basal 0.648 0.106 0.75 1.0 0.737 DNAH7 BRCA1 NF1 CREBBP Basal 0.648 0.213 0.667 1.0 0.714 SYNE1 BRCA1 NF1 LumA 0.737 0.1 1.0 1.0 0.847 TP53 CDH1 LumA 0.72 0.04 1.0 1.0 0.838 TP53 FOXA1 LumA 0.703 0.259 0.667 1.0 0.888 TP53 CDH1 CTCF LumA 0.7 0.109 1.0 1.0 0.691 TP53 NF1 LumA 0.696 0.033 1.0 0.792 0.959 TP53 AKT1 CDH1 LumA 0.686 0.195 0.75 0.894 0.906 TP53 ZFHX3 CDH1 CTCF LumA 0.679 0.145 1.0 0.632 0.939 TP53 CDH1 PTEN LumA 0.678 0.107 0.667 1.0 0.94 PIK3R1 AKT1 PTEN LumA 0.665 0.192 0.75 0.752 0.964 TP53 CDH1 PTEN CTCF LumA 0.659 0.202 0.667 1.0 0.768 TP53 NF1 SF3B1 LumA 0.651 0.13 0.75 0.75 0.974 TP53 AKT1 CDH1 CTCF LumA 0.65 0.043 0.667 1.0 0.891 AKT1 FOXA1 PTEN LumA 0.649 0.059 0.667 1.0 0.87 TP53 FOXA1 CTCF LumA 0.648 0.061 0.75 0.793 0.988 MAP3K1 TP53 AKT1 CDH1 LumA 0.642 0.243 0.667 0.809 0.849 TP53 FOXA1 GATA3 LumA 0.64 0.053 0.667 1.0 0.838 TP53 FOXA1 SF3B1 LumA 0.639 0.149 0.75 0.7 0.958 TP53 CDH1 PTEN GATA3 LumA 0.639 0.0 0.667 1.0 0.891 AKT1 PTEN MLLT4 LumA 0.633 0.076 0.667 1.0 0.789 FOXA1 PTEN CTCF LumA 0.633 0.136 0.75 0.663 0.981 MAP3K1 TP53 CDH1 PTEN LumA 0.627 0.177 0.5 0.9 0.93 TP53 CDH1 CTCF SF3B1

22 LumA 0.625 0.0 0.667 1.0 0.834 MED23 AKT1 PTEN LumA 0.62 0.157 0.75 0.581 0.99 PIK3R1 TP53 CDH1 PTEN LumA 0.619 0.13 0.5 0.895 0.953 TP53 CDH1 MLLT4 CTCF LumA 0.618 0.166 0.667 0.807 0.832 MAP3K1 TP53 NF1 LumA 0.617 0.156 0.5 0.95 0.862 TP53 CDH1 CSMD1 CTCF LumA 0.617 0.316 0.667 0.71 0.775 FOXA1 PTEN GATA3 LumA 0.617 0.163 0.5 0.837 0.97 PIK3R1 TP53 CDH1 CTCF LumA 0.617 0.223 0.75 0.567 0.929 FOXA1 CDH1 PTEN GATA3 LumA 0.617 0.0 0.667 0.872 0.93 AKT1 PTEN PIK3CA LumB 0.718 0.025 1.0 1.0 0.847 TP53 CDH1 LumB 0.718 0.008 1.0 0.904 0.959 TP53 AKT1 CDH1 LumB 0.702 0.066 1.0 1.0 0.743 CDH1 GATA3 LumB 0.691 0.098 1.0 1.0 0.665 PTEN GATA3 LumB 0.687 0.037 1.0 0.771 0.939 TP53 CDH1 PTEN LumB 0.67 0.104 0.667 1.0 0.911 TP53 HUWE1 CDH1 LumB 0.668 0.179 0.75 0.812 0.931 AXL TP53 CDH1 PTPRD LumB 0.667 0.0 1.0 1.0 0.67 ATM GATA3 LumB 0.664 0.087 0.75 0.873 0.945 AXL TP53 HUWE1 CDH1 LumB 0.662 0.093 0.75 0.863 0.942 FLNB AXL TP53 CDH1 LumB 0.661 0.047 0.75 0.876 0.973 TP53 AKT1 CDH1 PTPRD LumB 0.657 0.126 0.75 0.865 0.886 NF1 CDH1 PTEN GATA3 LumB 0.654 0.146 0.5 1.0 0.971 TP53 HUWE1 CDH1 ARID1A LumB 0.653 0.067 0.75 0.828 0.968 FLNB TP53 CDH1 PTEN LumB 0.652 0.115 0.667 0.904 0.924 TRRAP TP53 CDH1 LumB 0.649 0.111 0.75 0.782 0.953 AXL TRRAP TP53 CDH1 LumB 0.649 0.021 0.75 0.852 0.975 FLNB TP53 AKT1 CDH1 LumB 0.649 0.028 0.667 1.0 0.899 ATM CDH1 PTEN LumB 0.648 0.083 0.667 1.0 0.843 CDH1 PTEN GATA3 LumB 0.647 0.144 0.667 1.0 0.777 CDH1 PTPRD GATA3 LumB 0.646 0.127 0.667 1.0 0.79 HUWE1 CDH1 GATA3 LumB 0.643 0.109 0.75 0.785 0.926 AXL TP53 CDH1 GATA3 LumB 0.641 0.102 0.5 1.0 0.963 FLNB TP53 CDH1 ARID1A LumB 0.64 0.135 0.75 0.715 0.96 TP53 CDH1 PTEN PTPRD LumB 0.639 0.067 0.75 0.781 0.958 TP53 CDH1 PTEN GATA3 LumB 0.636 0.142 0.667 1.0 0.734 NF1 CDH1 GATA3 LumB 0.635 0.071 0.5 1.0 0.969 ERCC6 AKT1 CDH1 ARID1A LumB 0.632 0.022 0.75 0.803 0.952 MDN1 TP53 AKT1 CDH1 LumB 0.631 0.268 0.667 0.793 0.795 AXL TP53 PTPRD LumB 0.626 0.091 0.75 0.696 0.966 AXL TP53 CDH1 PTEN Her2 0.81 0.387 1.0 0.925 0.928 XPO1 TP53 BRCA2 Her2 0.783 0.273 1.0 0.875 0.983 XPO1 TP53 BRCA2 EGFR Her2 0.764 0.4 0.75 0.953 0.952 XPO1 SMC4 TP53 BRCA2 Her2 0.757 0.141 1.0 0.925 0.96 TP53 BRCA2 EGFR Her2 0.755 0.13 1.0 1.0 0.89 TP53 BRCA2 Her2 0.75 0.424 0.75 0.875 0.952 CASC5 XPO1 TP53 BRCA2 Her2 0.742 0.395 0.75 0.875 0.949 DSP XPO1 TP53 BRCA2 Her2 0.728 0.078 1.0 0.853 0.982 TYK2 TP53 EGFR CDH1 Her2 0.723 0.18 1.0 0.769 0.942 JAK2 TP53 CDH1 Her2 0.721 0.156 1.0 0.736 0.993 JAK2 TP53 EGFR CDH1 Her2 0.72 0.377 0.75 0.781 0.971 DLG1 XPO1 TP53 BRCA2 Her2 0.717 0.065 1.0 0.879 0.924 TYK2 TP53 EGFR Her2 0.717 0.052 1.0 0.925 0.891 TP53 BRCA2 PAX5 Her2 0.707 0.212 0.75 0.892 0.976 XPO1 TP53 BRCA2 CDH1 Her2 0.707 0.135 1.0 0.714 0.977 JAK2 TYK2 TP53 CDH1

23 Her2 0.707 0.103 1.0 0.769 0.955 TP53 EGFR CDH1 Her2 0.702 0.078 1.0 0.783 0.946 TP53 ZFHX3 EGFR CDH1 Her2 0.702 0.001 1.0 0.908 0.898 TYK2 TP53 CDH1 Her2 0.701 0.291 0.75 0.846 0.918 XPO1 TP53 BRCA2 FLG Her2 0.701 0.129 0.75 0.946 0.978 TP53 BRCA2 MCM6 EGFR Her2 0.696 0.302 0.667 0.925 0.89 CASC5 TP53 BRCA2 Her2 0.692 0.0 1.0 1.0 0.769 TYK2 TP53 Her2 0.691 0.106 0.75 0.946 0.962 IPO8 TP53 BRCA2 EGFR Her2 0.69 0.182 0.75 0.853 0.975 TP53 EGFR CDH1 CUL4B Her2 0.689 0.224 0.75 0.819 0.963 XPO1 TP53 EGFR CUL4B Her2 0.687 0.175 0.75 0.853 0.972 JAK2 TP53 CDH1 CUL4B Her2 0.687 0.168 0.667 1.0 0.912 TP53 BRCA2 TBL1XR1 Her2 0.686 0.113 0.75 0.892 0.991 TP53 BRCA2 EGFR CDH1 Her2 0.677 0.12 0.75 0.859 0.978 JAK2 TYK2 TP53 BRCA2 Her2 0.675 0.158 0.75 0.806 0.987 JAK2 TP53 BRCA2 CDH1

Table S9: Top driver modules inferred in receptors-classiﬁed breast cancer subtypes. Her2 refers to Her2- enriched. subtype score co-expression co-regulation ME PPI Gene1 Gene2 Gene3 Gene4 TN 0.708 0.118 0.75 1.0 0.964 MAP1A RB1 BRCA1 BRCA2 TN 0.706 0.127 1.0 0.807 0.89 TP53 BRCA2 TN 0.679 0.213 0.667 0.886 0.952 TAF1 TP53 BRCA2 TN 0.666 0.025 0.75 1.0 0.888 RB1 BRCA1 BRCA2 MUC5B TN 0.662 0.018 1.0 0.684 0.947 TP53 BRCA2 PTEN TN 0.659 0.0 1.0 0.684 0.953 RB1 TP53 BRCA2 TN 0.657 0.011 0.75 1.0 0.867 RB1 BRCA1 MUC5B PTEN TN 0.653 0.104 0.75 0.772 0.986 TAF1 TP53 BRCA2 PTEN TN 0.653 0.009 0.75 1.0 0.855 RB1 DYNC2H1 BRCA2 PTEN TN 0.651 0.009 1.0 0.611 0.985 RB1 TP53 BRCA2 PTEN TN 0.649 0.085 0.75 0.772 0.991 RB1 TAF1 TP53 BRCA2 TN 0.632 0.297 0.5 1.0 0.732 NID1 MUC16 PTEN VCAN TN 0.63 0.089 0.75 0.698 0.981 TP53 BRCA2 PTEN ARID1A TN 0.629 0.009 0.75 0.772 0.984 TP53 GRIN2B BRCA2 PTEN TN 0.628 0.107 0.667 0.798 0.941 TP53 BRCA2 ARID1A TN 0.626 0.093 0.75 0.698 0.962 DSP TP53 BRCA2 PTEN TN 0.625 0.151 0.75 0.696 0.901 TAF1 TP53 BRCA2 VCAN TN 0.625 0.085 0.75 0.671 0.992 TAF1 TP53 BRCA1 BRCA2 TN 0.624 0.176 0.5 0.861 0.961 DSP TAF1 TP53 BRCA2 TN 0.622 0.054 0.75 0.698 0.987 RB1 TP53 BRCA2 ARID1A TN 0.621 0.051 1.0 0.53 0.902 RB1 TP53 PTEN VCAN TN 0.617 0.058 0.75 0.698 0.964 DSP RB1 TP53 BRCA2 TN 0.615 0.074 0.75 0.653 0.984 TP53 BRCA2 PTEN CUL7 TN 0.614 0.102 1.0 0.578 0.774 TP53 PTEN VCAN TN 0.609 0.009 0.75 0.698 0.979 CDC42BPA TP53 BRCA2 PTEN TN 0.607 0.0 0.75 0.698 0.979 CDC42BPA RB1 TP53 BRCA2 TN 0.607 0.071 0.667 0.738 0.952 TP53 BRCA2 CUL7 TN 0.601 0.0 0.667 0.798 0.939 TP53 GRIN2B BRCA2 TN 0.597 0.0 0.667 0.798 0.924 TP53 BRCA2 CDC42BPA TN 0.594 0.085 0.5 0.861 0.932 TAF1 TP53 SPTB BRCA2 LumA 0.726 0.211 1.0 1.0 0.692 ERBB2 GATA3 LumA 0.725 0.0 1.0 1.0 0.9 TP53 ATM

24 LumA 0.722 0.0 1.0 1.0 0.89 TP53 BRCA2 LumA 0.711 0.226 1.0 0.792 0.825 ERBB2 MYB GATA3 LumA 0.674 0.206 0.75 0.847 0.894 FLNC ERBB2 MYB GATA3 LumA 0.653 0.185 0.667 1.0 0.759 FLNC ERBB2 GATA3 LumA 0.648 0.211 0.75 0.713 0.919 ERBB2 MYB GATA3 ARID1A LumA 0.634 0.105 1.0 0.687 0.743 CDH1 GATA3 LumA 0.633 0.14 0.667 0.888 0.839 ERBB2 PTEN GATA3 LumA 0.63 0.24 0.5 0.917 0.861 WDFY3 TP53 NF1 ARID1A LumA 0.627 0.095 0.5 0.947 0.965 TP53 ATM CDH1 CTCF LumA 0.627 0.0 1.0 0.817 0.691 TP53 NF1 LumA 0.623 0.115 0.667 0.888 0.822 TP53 NF1 ARID1A LumA 0.623 0.302 0.667 0.855 0.667 NF1 FOXA1 GATA3 LumA 0.617 0.212 0.667 0.844 0.744 ERBB2 NF1 GATA3 LumA 0.616 0.212 0.667 0.881 0.706 WDFY3 TP53 NF1 LumA 0.612 0.07 0.667 1.0 0.713 ERBB2 MYO7A GATA3 LumA 0.61 0.059 0.667 0.874 0.84 TP53 NF1 ERBB3 LumA 0.609 0.106 0.667 0.831 0.832 MAP3K1 TP53 NF1 LumA 0.606 0.146 0.5 0.867 0.912 TP53 NF1 CDH1 CTCF LumA 0.604 0.27 0.5 0.809 0.837 WDFY3 TP53 NF1 AHNAK LumA 0.604 0.133 0.5 0.877 0.905 TP53 NF1 CTCF ARID1A LumA 0.601 0.268 0.667 0.644 0.827 FOXA1 CDH1 GATA3 LumA 0.599 0.134 0.5 0.812 0.951 TP53 FOXA1 CDH1 GATA3 LumA 0.597 0.175 0.667 0.69 0.854 CDH1 MYB GATA3 LumA 0.594 0.116 0.5 0.823 0.937 TP53 CDH1 GATA3 CTCF LumA 0.593 0.231 0.667 0.685 0.79 CDH1 GATA3 CTCF LumA 0.592 0.166 0.5 0.762 0.941 TP53 ERBB2 FOXA1 GATA3 LumA 0.59 0.158 0.5 0.853 0.848 WDFY3 TP53 NF1 FOXA1 LumA 0.589 0.069 0.667 0.838 0.784 TP53 NF1 CTCF LumB 0.725 0.14 1.0 0.791 0.97 TP53 BRCA1 BRCA2 LumB 0.722 0.0 1.0 1.0 0.89 TP53 BRCA2 LumB 0.717 0.0 1.0 1.0 0.87 TP53 PTEN LumB 0.701 0.184 0.75 0.919 0.95 TP53 ERBB2 MYB GATA3 LumB 0.696 0.099 0.75 1.0 0.936 TP53 BRCA1 NF1 FOXA1 LumB 0.694 0.239 0.75 0.866 0.922 TP53 BRCA1 BRCA2 ASPM LumB 0.694 0.214 0.75 0.855 0.957 TP53 BRCA1 BRCA2 NF1 LumB 0.675 0.07 0.75 0.889 0.989 TP53 BRCA1 BRCA2 NCOR1 LumB 0.667 0.142 0.667 1.0 0.859 TP53 BRCA1 NF1 LumB 0.667 0.139 0.75 0.829 0.95 SPTBN2 TP53 BRCA1 BRCA2 LumB 0.666 0.184 0.75 0.783 0.947 ERBB2 PIK3CA MYB GATA3 LumB 0.666 0.145 0.667 1.0 0.851 TP53 BRCA2 NF1 LumB 0.666 0.056 0.667 1.0 0.941 TP53 BRCA1 ERBB2 LumB 0.659 0.032 0.667 1.0 0.938 TP53 BRCA1 FOXA1 LumB 0.655 0.07 0.75 0.829 0.971 MAP1A TP53 BRCA1 BRCA2 LumB 0.655 0.184 0.75 0.792 0.896 ERBB2 MYB GATA3 TTN LumB 0.649 0.07 0.75 0.783 0.994 TP53 BRCA1 BRCA2 ATM LumB 0.648 0.085 0.667 1.0 0.84 TP53 NF1 ERBB3 LumB 0.644 0.139 0.667 0.91 0.85 TP53 MYB GATA3 LumB 0.641 0.108 0.5 1.0 0.957 PIK3R1 TP53 BRCA1 NF1 LumB 0.64 0.088 0.75 0.806 0.916 TP53 BRCA2 ASPM ATM LumB 0.636 0.081 0.5 1.0 0.963 NCOA3 GRIK2 BRCA2 ATM LumB 0.636 0.055 0.667 1.0 0.822 TP53 NF1 ARID1A LumB 0.635 0.08 0.5 1.0 0.959 TP53 ERBB2 ITPR1 MYB LumB 0.633 0.123 0.5 1.0 0.909 TP53 BRCA2 GRIA1 NF1 LumB 0.632 0.091 0.5 1.0 0.937 TP53 SPTB BRCA1 ERBB2

25 LumB 0.631 0.002 0.667 1.0 0.855 TP53 ERBB2 NF1 LumB 0.63 0.114 0.5 1.0 0.905 ZFP36L1 TP53 BRCA2 NF1 LumB 0.626 0.0 0.667 0.879 0.959 TP53 BRCA2 ATM LumB 0.626 0.127 0.5 1.0 0.878 TP53 SPTB BRCA1 NF1 Her2 0.636 0.307 0.5 0.918 0.819 ZFP36L1 TP53 RUNX1 VCAN Her2 0.625 0.208 0.5 0.918 0.876 NCOA3 TP53 RUNX1 VCAN Her2 0.618 0.227 0.667 0.882 0.698 TP53 RUNX1 VCAN Her2 0.614 0.165 0.5 0.918 0.873 TP53 CDH1 RUNX1 VCAN Her2 0.613 0.18 0.5 0.918 0.853 NCOA3 ZFP36L1 TP53 VCAN Her2 0.611 0.097 0.5 0.878 0.969 NCOA3 TP53 SCN5A ERBB2 Her2 0.607 0.078 0.5 0.878 0.973 NCOA3 TP53 ERBB2 RUNX1 Her2 0.603 0.247 0.5 0.83 0.837 TP53 RUNX1 MYH9 VCAN Her2 0.602 0.094 0.667 0.882 0.765 NCOA3 TP53 VCAN Her2 0.602 0.129 0.5 0.918 0.861 NCOA3 TP53 SCN5A VCAN Her2 0.6 0.058 0.5 0.878 0.963 NCOA3 ZFP36L1 TP53 ERBB2 Her2 0.598 0.049 0.5 0.878 0.967 NCOA3 MTOR TP53 ERBB2 Her2 0.597 0.3 0.5 0.759 0.829 TP53 AHNAK RUNX1 VCAN Her2 0.593 0.007 0.5 0.878 0.987 NCOA3 TP53 GRIN2B ERBB2 Her2 0.593 0.165 0.667 0.882 0.659 ZFP36L1 TP53 VCAN Her2 0.592 0.101 0.5 0.918 0.849 ZFP36L1 TP53 CDH1 VCAN Her2 0.591 0.15 0.5 0.918 0.795 ZFP36L1 TP53 SCN5A VCAN Her2 0.589 0.19 0.5 0.759 0.908 NCOA3 DCC TP53 VCAN Her2 0.587 0.035 0.667 0.882 0.763 TP53 CDH1 VCAN Her2 0.587 0.211 0.5 0.759 0.877 NCOA3 TP53 AHNAK VCAN Her2 0.566 0.047 0.667 0.882 0.67 TP53 SCN5A VCAN Her2 0.563 0.076 0.5 0.759 0.915 NCOA3 TP53 ERBB2 VCAN Her2 0.557 0.143 0.5 0.759 0.824 HECW1 TP53 CDH1 VCAN Her2 0.555 0.181 0.667 0.882 0.492 TP53 MYH8 VCAN Her2 0.545 0.0 0.667 0.882 0.63 SYNE1 TP53 VCAN Her2 0.543 0.0 0.667 0.882 0.624 SMG1 TP53 VCAN Her2 0.54 0.062 0.667 0.882 0.552 TP53 ASPM VCAN Her2 0.539 0.209 0.0 1.0 0.948 NCOA3 ZFP36L1 RUNX1 TP53 Her2 0.537 0.0 0.667 0.882 0.6 TP53 WNK1 VCAN Her2 0.532 0.062 0.667 0.832 0.567 TP53 HRNR VCAN

26 3.6 Demonstrating the generality of ModulOmics via complexes detection We explored a proof-of-concept example for generalizing ModulOmics beyond identifying cancer driver modules. To this end, we collected known biological complexes from CORUM [17], resulting in 432 complexes of size 2, 501 complexes of size 3, and 179 complexes of size 5, and matched these with random groups of genes of the same size. For both the true complexes and the false random groups, we calculated the ModulOmics score using the breast cancer dataset [2], since it represented the largest cohort used in this study. biggest cohort of cancer expression - the breast cancer study, the HIPPIE PPI network and the TRRUST regulatory network, as they were used for detection of cancer modules. Using the sklearn Python package, we trained a logistic regression classifier to distinguish between true complexes and random groups, using the sklearn default suggested parameters (L2 regularization with C=1.0 factor to balance overfitting). The resulted classifier distinguished between the instances with an AUPR of 0.84 across all complex sizes S14, and with an AUPR of 0.9 when focusing only on modules of size 4, suggesting that bigger modules are more distinct that randomly generated groups of same size.

1.0

0.8

0.6

Precision 0.4

0.2 random complexes of size 4 AUPR (0.90) complexes AUPR (0.84) 0.0 0.0 0.2 0.4 0.6 0.8 1.0 Recall

Figure S14: Precession recall results for classifying known biological complexes and random group of genes, showing the classiﬁcation of 1, 112 known complexes (blue curve) and a restricted version of 179 known complexes of size 4 (pink curve).

27 References

[1] Omer An, Giovanni M. Dall’Olio, Thanos P. Mourikis, and Francesca D. Ciccarelli. Ncg 5.0: updates of a manually curated repository of cancer genes and associated properties from cancer mutational screenings. Nucleic Acids Research, 44(D1):D992–D999, 2016. [2] Cancer Genome Atlas Network. Comprehensive molecular portraits of human breast tumours. Nature, 490(7418):61–70, 2012.

[3] Cancer Genome Atlas Research Network. Comprehensive genomic characterization deﬁnes human glioblastoma genes and core pathways. Nature, 455(7216):1061–1068, October 2008. [4] Cancer Genome Atlas Research Network. Integrated genomic analyses of ovarian carcinoma. Nature, 474(7353):609–615, June 2011.

[5] Ethan Cerami, Jianjiong Gao, Ugur Dogrusoz, Benjamin E Gross, Selcuk Onur Sumer, Bülent Arman Aksoy, Anders Jacobsen, Caitlin J Byrne, Michael L Heuer, Erik Larsson, et al. The cbio cancer genomics portal: an open platform for exploring multidimensional cancer genomics data. Cancer Discovery, 2(5):401–404, 2012. [6] Simona Constantinescu, Ewa Szczurek, Pejman Mohammadi, JörgRahnenführer,and Niko Beeren- winkel. Timex: a waiting time model for mutually exclusive cancer alterations. Bioinformatics, 32(7):968–975, 2015. [7] Teresa Davoli, Andrew Wei W. Xu, Kristen E. Mengwasser, Laura M. Sack, John C. Yoon, Peter J. Park, and Stephen J. Elledge. Cumulative haploinsufficiency and triplosensitivity drive aneuploidy patterns and shape the cancer genome. Cell, 155(4):948–962, November 2013.

[8] Tamar Geiger, Juergen Cox, Pawel Ostasiewicz, Jacek R. Wisniewski, and Matthias Mann. Super- SILAC mix for quantitative proteomics of human tumor tissue. Nature Methods, 7(5):383–385, May 2010. [9] Heonjong Han, Hongseok Shim, Donghyun Shin, Jung Eun Shim, Yunhee Ko, Junha Shin, Hanhae Kim, Ara Cho, Eiru Kim, Tak Lee, et al. Trrust: a reference database of human transcriptional regulatory interactions. Scientiﬁc Reports, 5:11432, 2015. [10] Matan Hofree, Hannah Carter, Jason F Kreisberg, Sourav Bandyopadhyay, Paul S Mischel, Stephen Friend, and Trey Ideker. Challenges in identifying cancer genes by analysis of exome sequencing data. Nature Communications, 7, 2016. [11] J. L. Huret, S. Senon, A. Bernheim, and P. Dessen. An atlas on genes and chromosomes in oncology and haematology. Cellular and molecular biology, 50(7):805–807, November 2004. [12] Yoo-Ah Kim, Dong-Yeon Cho, Phuong Dao, and Teresa M Przytycka. Memcover: integrated analysis of mutual exclusivity and functional network reveals dysregulated pathways across multiple cancer types. Bioinformatics, 31(12):i284–i292, 2015. [13] Mark D. Leiserson, Fabio Vandin, Hsin-Ta T. Wu, Jason R. Dobson, Jonathan V. Eldridge, Jacob L. Thomas, Alexandra Papoutsaki, Younhun Kim, Beifang Niu, Michael McLellan, Michael S. Lawrence, Abel Gonzalez-Perez, David Tamborero, Yuwei Cheng, Gregory A. Ryslik, Nuria Lopez-Bigas, Gad Getz, Li Ding, and Benjamin J. Raphael. Pan-cancer network analysis identiﬁes combinations of rare somatic mutations across pathways and protein complexes. Nature Genetics, 47(2):106–114, February 2015.

[14] Evan O. Paull, Daniel E. Carlin, Mario Niepel, Peter K. Sorger, David Haussler, and Joshua M. Stu- art. Discovering causal pathways linking genomic events to transcriptional states using Tied Diﬀusion Through Interacting Events (TieDIE). Bioinformatics, 29(21):2757–2764, November 2013.

28 [15] Sune Pletscher-Frankild, Albert Pallej`a,Kalliopi Tsafou, Janos X. Binder, and Lars Juhl J. Jensen. DISEASES: text mining and data integration of disease-gene associations. Methods, 74:83–89, March 2015. [16] Yair Pozniak, Nora Balint-Lahat, JanDaniel Rudolph, Cecilia Lindskog, Rotem Katzir, Camilla Avivi, Fredrik Pontn, Eytan Ruppin, Iris Barshack, and Tamar Geiger. System-wide clinical proteomics of breast cancer reveals global remodeling of tissue homeostasis. Cell Systems, 2(3):172 – 184, 2016. [17] Andreas Ruepp, Brigitte Waegele, Martin Lechner, Barbara Brauner, Irmtraud Dunger-Kaltenbach, Gisela Fobo, Goar Frishman, Corinna Montrone, and H-Werner W. Mewes. CORUM: the comprehensive resource of mammalian protein complexes–2009. Nucleic acids research, 38(Database issue):D497–501, January 2010.

[18] Martin H. Schaefer, Jean-Fred F. Fontaine, Arunachalam Vinayagam, Pablo Porras, Erich E. Wanker, and Miguel A. Andrade-Navarro. HIPPIE: Integrating protein interaction networks with experiment based quality scores. PLoS ONE, 7(2), 2012. [19] Aravind Subramanian, Pablo Tamayo, Vamsi K. Mootha, Sayan Mukherjee, Benjamin L. Ebert, Michael A. Gillette, Amanda Paulovich, Scott L. Pomeroy, Todd R. Golub, Eric S. Lander, and Jill P. Mesirov. Gene set enrichment analysis: A knowledge-based approach for interpreting genome-wide expression proﬁles. Proceedings of the National Academy of Sciences, 102(43):15545–15550, October 2005. [20] Stefka Tyanova, Reidar Albrechtsen, Pauliina Kronqvist, Juergen Cox, Matthias Mann, and Tamar Geiger. Proteomic maps of breast cancer subtypes. Nature Communications, 7:10259, 2016.

[21] Igor Ulitsky, Adi Maron-Katz, Seagull Shavit, Dorit Sagir, Chaim Linhart, Ran Elkon, Amos Tanay, Roded Sharan, Yosef Shiloh, and Ron Shamir. Expander: from expression microarrays to networks and functions. Nature Protocols, 5(2):303–322, February 2010. [22] UniProt Consortium. UniProt: a hub for protein information. Nucleic acids research, 43(Database issue):D204–D212, January 2015.