Glioblastoma Multiforme: Mutation Analysis (MutSig)
Maintained by Dan DiCara (Broad Institute)
Overview
Introduction

This report serves to describe the mutational landscape and properties of a given individual set, as well as rank genes and genesets according to mutational significance. MutSig v2.0 was used to generate the results found in this report.

Working with individual set: GBM.

Number of patients in set: 276

Input

The input for this pipeline is a set of individuals with the following files associated for each:

1. An annotated .maf file describing the mutations called for the respective individual, and their properties.

2. A .wig file that contains information about the coverage of the sample.

Summary

Significantly mutated genes (q ≤ 0.1): 223

Mutations seen in COSMIC: 445

Significantly mutated genes in COSMIC territory: 55

Genes with clustered mutations (&le 3 aa apart): 385

Significantly mutated genesets: 170

Significantly mutated genesets: (excluding sig. mutated genes): 0

Results
Breakdown of Mutations by Type

Table 1.  Get Full Table Table representing breakdown of mutations by type.

type count
De_novo_Start_InFrame 6
De_novo_Start_OutOfFrame 10
Frame_Shift_Del 812
Frame_Shift_Ins 362
In_Frame_Del 328
In_Frame_Ins 36
Indel 1
Missense_Mutation 23424
Nonsense_Mutation 1745
Nonstop_Mutation 17
Silent 8876
Splice_Site_DNP 11
Splice_Site_Del 101
Splice_Site_Ins 28
Splice_Site_ONP 1
Splice_Site_SNP 469
Start_Codon_Del 2
Start_Codon_Ins 1
Stop_Codon_Del 2
Total 36232
Breakdown of Mutation Rates by Category Type

Table 2.  Get Full Table A breakdown of mutation rates per category discovered for this individual set.

category n N rate rate_per_mb relative_rate
*CpG->T 8794 406970401 0.000022 22 6
*Cp(A/C/T)->T 3993 3448378213 1.2e-06 1.2 0.32
C->(G/A) 5881 3855348614 1.5e-06 1.5 0.43
A->mut 4753 3770634164 1.3e-06 1.3 0.35
indel+null 3856 7625982864 5.1e-07 0.51 0.14
double_null 79 7625982864 1e-08 0.01 0.0029
Total 27356 7625982864 3.6e-06 3.6 1
Target Coverage for Each Individual

The x axis represents the samples. The y axis represents the exons, one row per exon, and they are sorted by average coverage across samples. For exons with exactly the same average coverage, they are sorted next by the %GC of the exon. (The secondary sort is especially useful for the zero-coverage exons at the bottom).

Figure 1. 

Distribution of Mutation Counts, Coverage, and Mutation Rates Across Samples

Figure 2. 

CoMut Plot

Figure 3.  Get High-res Image The matrix in the center of the figure represents individual mutations in patient samples, color-coded by type of mutation, for the significantly mutated genes. The rate of synonymous and non-synonymous mutations is displayed at the top of the matrix. The barplot on the left of the matrix shows the number of mutations in each gene. The percentages represent the fraction of tumors with at least one mutation in the specified gene. The barplot to the right of the matrix displays the q-values for the most significantly mutated genes. The purple boxplots below the matrix (only displayed if required columns are present in the provided MAF) represent the distributions of allelic fractions observed in each sample. The plot at the bottom represents the base substitution distribution of individual samples, using the same categories that were used to calculate significance.

Significantly Mutated Genes

Table 3.  Get Full Table A Ranked List of Significantly Mutated Genes. Number of significant genes found: 223. Number of genes displayed: 35

rank gene description N n npat nsite nsil n1 n2 n3 n4 n5 n6 p_classic p_ks p_cons p_joint p q
1 PIK3R1 phosphoinositide-3-kinase, regulatory subunit 1 (alpha) 642722 17 17 14 0 0 3 2 4 8 0 <1.00e-15 0.000013 0.0044 0.000018 <0.00 <0.00
2 IDH1 isocitrate dehydrogenase 1 (NADP+), soluble 349470 15 15 2 0 14 0 1 0 0 0 1.11e-15 2e-07 0.97 0.000011 0.00 0.00
3 KRTAP5-5 keratin associated protein 5-5 181491 18 16 7 0 0 0 3 1 14 0 1.22e-15 0.00082 1 0.0023 1.11e-16 6.61e-13
4 PIK3CA phosphoinositide-3-kinase, catalytic, alpha polypeptide 885302 19 19 17 1 3 5 2 7 2 0 1.24e-14 0.015 0.0048 0.00068 3.33e-16 1.12e-12
5 TP53 tumor protein p53 344937 89 80 63 1 23 15 12 13 24 2 <1.00e-15 2e-07 2e-07 0 <1.00e-15 <1.12e-12
6 EGFR epidermal growth factor receptor (erythroblastic leukemia viral (v-erb-b) oncogene homolog, avian) 1089173 58 54 26 5 9 21 23 5 0 0 <1.00e-15 2e-07 2e-07 0 <1.00e-15 <1.12e-12
7 MICALCL MICAL C-terminal like 537385 13 13 3 0 0 0 0 1 12 0 1.55e-13 2e-07 0.6 0 <1.00e-15 <1.12e-12
8 C7orf52 chromosome 7 open reading frame 52 247344 6 6 2 1 1 5 0 0 0 0 0.00043 0.0017 4e-07 0 <1.00e-15 <1.12e-12
9 KDSR 3-ketodihydrosphingosine reductase 270146 5 5 1 0 0 0 5 0 0 0 0.000095 2e-07 0.63 0 <1.00e-15 <1.12e-12
10 PLA2G2A phospholipase A2, group IIA (platelets, synovial fluid) 123007 5 5 1 0 0 0 5 0 0 0 2.64e-06 2e-07 0.25 0 <1.00e-15 <1.12e-12
11 MRM1 mitochondrial rRNA methyltransferase 1 homolog (S. cerevisiae) 284558 4 4 1 0 0 0 0 4 0 0 0.0016 4e-07 0.39 0 <1.00e-15 <1.12e-12
12 CD4 CD4 molecule 379108 3 3 1 1 0 0 0 3 0 0 0.015 3.8e-06 0.00045 0 <1.00e-15 <1.12e-12
13 ACSM2B acyl-CoA synthetase medium-chain family member 2B 454814 2 2 2 2 0 0 0 1 1 0 0.11 0.00024 0.68 0 <1.00e-15 <1.12e-12
14 C1orf100 chromosome 1 open reading frame 100 126907 2 2 2 0 0 0 0 0 2 0 0.036 0.013 0.75 0 <1.00e-15 <1.12e-12
15 TAPBPL TAP binding protein-like 353808 2 2 2 1 0 1 0 0 1 0 0.22 0.00055 0.68 0 <1.00e-15 <1.12e-12
16 ABCA3 ATP-binding cassette, sub-family A (ABC1), member 3 1218196 3 1 3 1 0 0 2 1 0 0 0.94 6e-07 0.012 0 <1.00e-15 <1.12e-12
17 PTEN phosphatase and tensin homolog (mutated in multiple advanced cancers 1) 314597 80 79 65 0 7 19 8 11 35 0 <1.00e-15 0.039 0.38 0.064 <2.44e-15 <2.56e-12
18 RB1 retinoblastoma 1 (including osteosarcoma) 722445 21 21 19 1 1 1 1 0 18 0 <1.00e-15 0.064 0.38 0.12 <4.55e-15 <4.51e-12
19 FOXRED1 FAD-dependent oxidoreductase domain containing 1 345284 9 9 1 0 0 0 0 9 0 0 1.92e-09 2e-07 0.93 0.000019 1.18e-12 1.11e-09
20 NF1 neurofibromin 1 (neurofibromatosis, von Recklinghausen disease, Watson disease) 2380799 31 25 30 1 2 5 3 4 14 3 8.04e-14 0.47 0.71 0.64 1.63e-12 1.45e-09
21 SPTA1 spectrin, alpha, erythrocytic 1 (elliptocytosis 2) 2040988 30 26 28 5 12 4 6 1 7 0 5.42e-13 0.27 0.29 0.37 6.05e-12 5.14e-09
22 COMMD4 COMM domain containing 4 147026 6 6 1 0 0 0 6 0 0 0 2.28e-07 2e-07 1 0.000015 9.26e-11 7.32e-08
23 CRYBA2 crystallin, beta A2 136089 6 6 3 1 0 0 5 1 0 0 1.83e-07 2e-07 0.86 0.000019 9.43e-11 7.32e-08
24 RYR2 ryanodine receptor 2 (cardiac) 3381004 36 28 35 7 21 3 6 6 0 0 3.66e-11 0.11 0.37 0.14 1.42e-10 1.06e-07
25 NRBP2 nuclear receptor binding protein 2 119779 4 4 1 0 0 4 0 0 0 0 0.000072 4e-07 0.99 6e-07 1.08e-09 7.70e-07
26 FUT7 fucosyltransferase 7 (alpha (1,3) fucosyltransferase) 113313 5 5 1 0 0 0 5 0 0 0 4.02e-06 2e-07 0.97 0.000015 1.50e-09 1.03e-06
27 HMP19 143577 5 5 1 0 0 0 5 0 0 0 8.41e-06 2e-07 0.93 0.000011 2.23e-09 1.47e-06
28 MST1 macrophage stimulating 1 (hepatocyte growth factor-like) 467340 10 9 6 0 2 0 0 1 7 0 6.93e-09 0.0053 0.89 0.014 2.35e-09 1.50e-06
29 ZNF704 zinc finger protein 704 335326 5 5 1 0 0 0 0 0 5 0 0.00014 4e-07 0.91 8e-07 2.70e-09 1.66e-06
30 FLG filaggrin 3183335 30 28 30 13 12 6 6 3 3 0 1.25e-09 0.6 0.044 0.13 3.94e-09 2.29e-06
31 EML2 echinoderm microtubule associated protein like 2 517871 8 8 3 1 0 0 0 0 8 0 4.45e-06 0.000014 0.82 0.000038 3.97e-09 2.29e-06
32 DNASE1L1 deoxyribonuclease I-like 1 189681 6 6 2 1 1 0 5 0 0 0 1.24e-06 0.0001 0.081 0.00014 4.15e-09 2.31e-06
33 FOXC1 forkhead box C1 130178 4 4 3 1 0 0 3 1 0 0 0.00026 2e-07 0.94 8e-07 4.92e-09 2.66e-06
34 TMEM45A transmembrane protein 45A 233464 5 5 1 0 0 0 0 5 0 0 0.000033 2e-07 0.08 7.2e-06 5.47e-09 2.87e-06
35 ZNF784 zinc finger protein 784 87173 3 3 1 0 0 0 0 0 3 0 0.00029 3.4e-06 0.42 3.6e-06 2.23e-08 0.000011

Note:

N - number of sequenced bases in this gene across the individual set.

n - number of (nonsilent) mutations in this gene across the individual set.

npat - number of patients (individuals) with at least one nonsilent mutation.

nsite - number of unique sites having a non-silent mutation.

nsil - number of silent mutations in this gene across the individual set.

n1 - number of nonsilent mutations of type: *CpG->T .

n2 - number of nonsilent mutations of type: *Cp(A/C/T)->T .

n3 - number of nonsilent mutations of type: C->(G/A) .

n4 - number of nonsilent mutations of type: A->mut .

n5 - number of nonsilent mutations of type: indel+null .

null - mutation category that includes nonsense, frameshift, splice-site mutations

p_classic = p-value for the observed amount of nonsilent mutations being elevated in this gene

p_ns_s = p-value for the observed nonsilent/silent ratio being elevated in this gene

p = p-value (overall)

q = q-value, False Discovery Rate (Benjamini-Hochberg procedure)

COSMIC analyses

In this analysis, COSMIC is used as a filter to increase power by restricting the territory of each gene. Cosmic version: v48.

Table 4.  Get Full Table Significantly mutated genes (COSMIC territory only). To access the database please go to: COSMIC. Number of significant genes found: 55. Number of genes displayed: 10

rank gene description n cos n_cos N_cos cos_ev p q
1 PTEN phosphatase and tensin homolog (mutated in multiple advanced cancers 1) 80 728 79 200928 2714 0 0
2 TP53 tumor protein p53 89 823 89 227148 26312 0 0
3 IDH1 isocitrate dehydrogenase 1 (NADP+), soluble 15 3 15 828 22380 2.9e-14 4.4e-11
4 PIK3R1 phosphoinositide-3-kinase, regulatory subunit 1 (alpha) 17 33 9 9108 20 3.1e-13 3.5e-10
5 PIK3CA phosphoinositide-3-kinase, catalytic, alpha polypeptide 19 184 15 50784 3911 1.5e-12 1.3e-09
6 EGFR epidermal growth factor receptor (erythroblastic leukemia viral (v-erb-b) oncogene homolog, avian) 58 218 46 60168 640 1.7e-12 1.3e-09
7 RB1 retinoblastoma 1 (including osteosarcoma) 21 267 11 73692 30 2e-12 1.3e-09
8 NF1 neurofibromin 1 (neurofibromatosis, von Recklinghausen disease, Watson disease) 31 285 10 78660 18 2.8e-12 1.6e-09
9 PTPN11 protein tyrosine phosphatase, non-receptor type 11 (Noonan syndrome 1) 6 31 4 8556 16 3.6e-08 0.000018
10 BRAF v-raf murine sarcoma viral oncogene homolog B1 6 88 5 24288 42058 3.9e-08 0.000018

Note:

n - number of (nonsilent) mutations in this gene across the individual set.

cos = number of unique mutated sites in this gene in COSMIC

n_cos = overlap between n and cos.

N_cos = number of individuals times cos.

cos_ev = total evidence: number of reports in COSMIC for mutations seen in this gene.

p = p-value for seeing the observed amount of overlap in this gene)

q = q-value, False Discovery Rate (Benjamini-Hochberg procedure)

Clustered Mutations

Table 5.  Get Full Table Genes with Clustered Mutations

num gene desc n mindist nmuts0 nmuts3 nmuts12 npairs0 npairs3 npairs12
2963 EGFR epidermal growth factor receptor (erythroblastic leukemia viral (v-erb-b) oncogene homolog, avian) 58 0 158 172 199 158 172 199
4499 IDH1 isocitrate dehydrogenase 1 (NADP+), soluble 15 0 105 105 105 105 105 105
10104 TP53 tumor protein p53 89 0 69 161 418 69 161 418
3606 FOXRED1 FAD-dependent oxidoreductase domain containing 1 9 0 36 36 36 36 36 36
7900 PTEN phosphatase and tensin homolog (mutated in multiple advanced cancers 1) 80 0 25 74 157 25 74 157
5436 LPHN1 latrophilin 1 10 0 16 16 18 16 16 18
878 BBS1 Bardet-Biedl syndrome 1 8 0 15 15 15 15 15 15
2202 COL5A3 collagen, type V, alpha 3 11 0 15 15 15 15 15 15
2215 COMMD4 COMM domain containing 4 6 0 15 15 15 15 15 15
2845 DSCAM Down syndrome cell adhesion molecule 10 0 15 15 15 15 15 15

Note:

n - number of mutations in this gene in the individual set.

mindist - distance (in aa) between closest pair of mutations in this gene

npairs3 - how many pairs of mutations are within 3 aa of each other.

npairs12 - how many pairs of mutations are within 12 aa of each other.

Geneset Analyses

Table 6.  Get Full Table A Ranked List of Significantly Mutated Genesets. (Source: MSigDB GSEA Cannonical Pathway Set).Number of significant genesets found: 170. Number of genesets displayed: 10

rank geneset description genes N_genes mut_tally N n npat nsite nsil n1 n2 n3 n4 n5 n6 p q
1 HSA04010_MAPK_SIGNALING_PATHWAY Genes involved in MAPK signaling pathway ACVR1B, ACVR1C, AKT1, AKT2, AKT3, ARRB1, ARRB2, ATF2, ATF4, BDNF, BRAF, CACNA1A, CACNA1B, CACNA1C, CACNA1D, CACNA1E, CACNA1F, CACNA1G, CACNA1H, CACNA1I, CACNA1S, CACNA2D1, CACNA2D2, CACNA2D3, CACNA2D4, CACNB1, CACNB2, CACNB3, CACNB4, CACNG1, CACNG2, CACNG3, CACNG4, CACNG5, CACNG6, CACNG7, CACNG8, CASP3, CD14, CDC25B, CDC42, CHP, CHUK, CRK, CRKL, DAXX, DDIT3, DUSP1, DUSP10, DUSP14, DUSP16, DUSP2, DUSP3, DUSP4, DUSP5, DUSP6, DUSP7, DUSP8, DUSP9, ECSIT, EGF, EGFR, ELK1, ELK4, EVI1, FAS, FASLG, FGF1, FGF10, FGF11, FGF12, FGF13, FGF14, FGF16, FGF17, FGF18, FGF19, FGF2, FGF20, FGF21, FGF22, FGF23, FGF3, FGF4, FGF5, FGF6, FGF7, FGF8, FGF9, FGFR1, FGFR2, FGFR3, FGFR4, FLNA, FLNB, FLNC, FOS, GADD45A, GADD45B, GADD45G, GNA12, GNG12, GRB2, HRAS, IKBKB, IKBKG, IL1A, IL1B, IL1R1, IL1R2, JUN, JUND, KRAS, LOC653852, MAP2K1, MAP2K1IP1, MAP2K2, MAP2K3, MAP2K4, MAP2K5, MAP2K6, MAP2K7, MAP3K1, MAP3K10, MAP3K12, MAP3K13, MAP3K14, MAP3K2, MAP3K3, MAP3K4, MAP3K5, MAP3K6, MAP3K7, MAP3K7IP1, MAP3K7IP2, MAP3K8, MAP4K1, MAP4K2, MAP4K3, MAP4K4, MAPK1, MAPK10, MAPK11, MAPK12, MAPK13, MAPK14, MAPK3, MAPK7, MAPK8, MAPK8IP1, MAPK8IP2, MAPK8IP3, MAPK9, MAPKAPK2, MAPKAPK3, MAPKAPK5, MAPT, MAX, MEF2C, MKNK1, MKNK2, MOS, MRAS, MYC, NF1, NFATC2, NFATC4, NFKB1, NFKB2, NGFB, NLK, NR4A1, NRAS, NTF3, NTF5, NTRK1, NTRK2, PAK1, PAK2, PDGFA, PDGFB, PDGFRA, PDGFRB, PLA2G10, PLA2G12A, PLA2G12B, PLA2G1B, PLA2G2A, PLA2G2D, PLA2G2E, PLA2G2F, PLA2G3, PLA2G4A, PLA2G5, PLA2G6, PPM1A, PPM1B, PPP3CA, PPP3CB, PPP3CC, PPP3R1, PPP3R2, PPP5C, PRKACA, PRKACB, PRKACG, PRKCA, PRKCB1, PRKCG, PRKX, PRKY, PTPN5, PTPN7, PTPRR, RAC1, RAC2, RAC3, RAF1, RAP1A, RAP1B, RAPGEF2, RASA1, RASA2, RASGRF1, RASGRF2, RASGRP1, RASGRP2, RASGRP3, RASGRP4, RPS6KA1, RPS6KA2, RPS6KA3, RPS6KA4, RPS6KA5, RPS6KA6, RRAS, RRAS2, SOS1, SOS2, SRF, STK3, STK4, STMN1, TAOK1, TAOK2, TAOK3, TGFB1, TGFB2, TGFB3, TGFBR1, TGFBR2, TNF, TNFRSF1A, TP53, TRAF2, TRAF6, ZAK 246 ACVR1C(2), AKT1(1), AKT2(1), AKT3(1), ATF2(1), ATF4(1), BDNF(1), BRAF(6), CACNA1A(4), CACNA1B(7), CACNA1C(5), CACNA1D(9), CACNA1E(9), CACNA1F(10), CACNA1G(6), CACNA1H(9), CACNA1I(5), CACNA1S(12), CACNA2D1(8), CACNA2D2(1), CACNA2D3(3), CACNA2D4(1), CACNB2(1), CACNB3(1), CACNG1(1), CACNG2(1), CACNG3(4), CACNG6(1), CACNG7(1), CACNG8(1), CD14(2), CDC25B(1), CHUK(4), DAXX(4), DDIT3(1), DUSP10(1), DUSP14(1), DUSP16(1), DUSP2(2), DUSP4(3), DUSP8(1), DUSP9(1), EGF(3), EGFR(58), ELK1(1), ELK4(1), FGF10(1), FGF13(1), FGF14(2), FGF18(1), FGF2(1), FGF20(2), FGF21(1), FGF23(3), FGF3(1), FGF6(3), FGF9(1), FGFR1(3), FGFR2(3), FGFR3(2), FGFR4(3), FLNA(14), FLNB(4), FLNC(6), FOS(1), GADD45B(1), GNA12(2), IKBKB(2), IL1R1(2), KRAS(1), MAP2K1(1), MAP2K2(1), MAP2K3(1), MAP2K5(1), MAP2K7(1), MAP3K1(5), MAP3K10(3), MAP3K12(5), MAP3K13(2), MAP3K14(2), MAP3K2(1), MAP3K3(4), MAP3K4(2), MAP3K5(1), MAP3K6(5), MAP3K7(2), MAP4K1(1), MAP4K2(1), MAP4K3(1), MAP4K4(5), MAPK1(3), MAPK13(1), MAPK14(1), MAPK3(2), MAPK7(3), MAPK8IP1(3), MAPK8IP2(3), MAPK8IP3(5), MAPK9(2), MAPKAPK2(1), MAPKAPK3(1), MAPT(1), MAX(2), MEF2C(1), MKNK1(1), NF1(31), NFATC2(1), NFATC4(2), NFKB1(1), NFKB2(1), NR4A1(3), NRAS(1), NTF3(1), NTRK1(4), NTRK2(1), PAK1(2), PDGFRA(10), PDGFRB(5), PLA2G12A(3), PLA2G2A(5), PLA2G2F(2), PLA2G3(2), PLA2G4A(4), PLA2G5(1), PLA2G6(3), PPM1B(1), PPP3CB(1), PPP3R1(1), PPP3R2(1), PPP5C(1), PRKACG(1), PRKCA(3), PRKCG(1), PRKX(2), PTPRR(2), RAP1B(2), RAPGEF2(4), RASA2(1), RASGRF1(5), RASGRF2(6), RASGRP1(1), RASGRP2(2), RASGRP3(2), RASGRP4(3), RPS6KA2(2), RPS6KA3(5), RRAS2(3), SOS1(1), SOS2(4), SRF(2), STK3(2), STK4(1), STMN1(1), TAOK1(1), TAOK2(2), TAOK3(2), TGFB1(1), TGFB2(1), TGFBR1(1), TGFBR2(5), TNFRSF1A(2), TP53(89), TRAF2(2), ZAK(1) 111569559 586 212 515 148 184 87 111 98 100 6 <1.00e-15 <1.28e-14
2 HSA04115_P53_SIGNALING_PATHWAY Genes involved in p53 signaling pathway APAF1, ATM, ATR, BAI1, BAX, BBC3, BID, CASP3, CASP8, CASP9, CCNB1, CCNB2, CCNB3, CCND1, CCND2, CCND3, CCNE1, CCNE2, CCNG1, CCNG2, CD82, CDC2, CDK2, CDK4, CDK6, CDKN1A, CDKN2A, CHEK1, CHEK2, CYCS, DDB2, EI24, FAS, GADD45A, GADD45B, GADD45G, GTSE1, IGF1, IGFBP3, LRDD, MDM2, MDM4, P53AIP1, PERP, PMAIP1, PPM1D, PTEN, RCHY1, RFWD2, RPRM, RRM2, RRM2B, SCOTIN, SERPINB5, SERPINE1, SESN1, SESN2, SESN3, SFN, SIAH1, STEAP3, THBS1, TNFRSF10B, TP53, TP53I3, TP73, TSC2, ZMAT3 64 APAF1(1), ATM(7), ATR(9), BAI1(1), BID(4), CCNB3(1), CCND2(1), CCNE1(1), CCNE2(1), CCNG1(2), CDKN1A(1), CDKN2A(3), CHEK1(5), DDB2(1), GADD45B(1), GTSE1(2), IGFBP3(1), MDM2(1), MDM4(2), PERP(3), PPM1D(1), PTEN(80), RFWD2(2), RPRM(1), SERPINB5(2), SERPINE1(1), SESN1(2), SESN2(1), SESN3(3), STEAP3(2), THBS1(3), TP53(89), TP53I3(2), TP73(1), TSC2(2) 25535861 240 158 196 28 44 51 32 39 71 3 <1.00e-15 <1.28e-14
3 ST_INTEGRIN_SIGNALING_PATHWAY Integrins are transmembrane receptors that mediate cell growth, survival, and migration by binding to ligands in the extracellular matrix. ABL1, ACK1, ACTN1, ACTR2, ACTR3, AKT1, AKT2, AKT3, ANGPTL2, ARHGEF6, ARHGEF7, BCAR1, BRAF, CAV1, CDC42, CDKN2A, CRK, CSE1L, DDEF1, DOCK1, EPHB2, FYN, GRAF, GRB2, GRB7, GRF2, GRLF1, ILK, ITGA1, ITGA10, ITGA11, ITGA2, ITGA3, ITGA4, ITGA5, ITGA6, ITGA7, ITGA8, ITGA9, ITGB3BP, MAP2K4, MAP2K7, MAP3K11, MAPK1, MAPK10, MAPK8, MAPK8IP1, MAPK8IP2, MAPK8IP3, MAPK9, MRAS, MYLK, MYLK2, P4HB, PAK1, PAK2, PAK3, PAK4, PAK6, PAK7, PIK3CA, PIK3CB, PKLR, PLCG1, PLCG2, PTEN, PTK2, RAF1, RALA, RHO, ROCK1, ROCK2, SHC1, SOS1, SOS2, SRC, TERF2IP, TLN1, TLN2, VASP, WAS, ZYX 78 ABL1(1), ACTN1(1), ACTR3(2), AKT1(1), AKT2(1), AKT3(1), ANGPTL2(2), ARHGEF7(1), BRAF(6), CAV1(1), CDKN2A(3), CSE1L(2), DOCK1(9), EPHB2(6), GRB7(1), GRLF1(4), ILK(2), ITGA1(1), ITGA10(3), ITGA11(2), ITGA2(4), ITGA3(2), ITGA4(7), ITGA5(1), ITGA6(2), ITGA7(3), ITGA8(7), ITGA9(1), MAP2K7(1), MAPK1(3), MAPK8IP1(3), MAPK8IP2(3), MAPK8IP3(5), MAPK9(2), MYLK(7), MYLK2(1), PAK1(2), PAK3(2), PAK6(1), PAK7(3), PIK3CA(19), PIK3CB(2), PKLR(4), PLCG1(3), PLCG2(11), PTEN(80), PTK2(2), RALA(1), RHO(1), ROCK1(3), ROCK2(4), SOS1(1), SOS2(4), SRC(2), TERF2IP(3), TLN1(6), TLN2(5), VASP(2), ZYX(2) 48062722 265 143 245 58 68 49 41 50 57 0 <1.00e-15 <1.28e-14
4 ST_FAS_SIGNALING_PATHWAY The Fas receptor induces apoptosis and NF-kB activation when bound to Fas ligand. ADPRT, ALG2, BAK1, BAX, BFAR, BIRC4, BTK, CAD, CASP10, CASP3, CASP8, CASP8AP2, CD7, CDK2AP1, CSNK1A1, DAXX, DEDD, DEDD2, DFFA, DIABLO, EGFR, EPHB2, FADD, FAF1, FAIM2, FREQ, HRB, HSPB1, IL1A, IL8, MAP2K4, MAP2K7, MAP3K1, MAP3K5, MAPK1, MAPK10, MAPK8, MAPK8IP1, MAPK8IP2, MAPK8IP3, MAPK9, MCP, MET, NFAT5, NFKB1, NFKB2, NFKBIA, NFKBIB, NFKBIE, NFKBIL1, NFKBIL2, NR0B2, PFN1, PFN2, PTPN13, RALBP1, RIPK1, ROCK1, SMPD1, TNFRSF6, TNFRSF6B, TP53, TPX2, TRAF2, TUFM, VIL2 59 BTK(1), CAD(6), CASP8AP2(3), CD7(2), DAXX(4), DEDD2(2), DFFA(1), DIABLO(1), EGFR(58), EPHB2(6), FAF1(3), FAIM2(1), MAP2K7(1), MAP3K1(5), MAP3K5(1), MAPK1(3), MAPK8IP1(3), MAPK8IP2(3), MAPK8IP3(5), MAPK9(2), MET(5), NFAT5(3), NFKB1(1), NFKB2(1), NFKBIA(2), NFKBIB(1), NFKBIE(1), NFKBIL1(1), NFKBIL2(6), NR0B2(3), PTPN13(4), RALBP1(3), RIPK1(1), ROCK1(3), TNFRSF6B(2), TP53(89), TPX2(1), TRAF2(2), TUFM(1) 28393004 241 142 182 36 57 53 55 36 38 2 <1.00e-15 <1.28e-14
5 SIG_CHEMOTAXIS Genes related to chemotaxis ACTR2, ACTR3, AKT1, AKT2, AKT3, ANGPTL2, ARHGAP1, ARHGAP4, ARHGEF11, BTK, CDC42, CFL1, CFL2, GDI1, GDI2, INPPL1, ITPR1, ITPR2, ITPR3, LIMK1, MYLK, MYLK2, P101-PI3K, PAK1, PAK2, PAK3, PAK4, PAK6, PAK7, PDK1, PIK3CA, PIK3CD, PIK3CG, PIK3R1, PITX2, PPP1R13B, PTEN, RACGAP1, RHO, ROCK1, ROCK2, RPS4X, SAG, WASF1, WASL 44 ACTR3(2), AKT1(1), AKT2(1), AKT3(1), ANGPTL2(2), ARHGAP1(1), ARHGAP4(1), ARHGEF11(3), BTK(1), CFL2(1), INPPL1(4), ITPR1(7), ITPR2(11), ITPR3(11), LIMK1(1), MYLK(7), MYLK2(1), PAK1(2), PAK3(2), PAK6(1), PAK7(3), PIK3CA(19), PIK3CD(2), PIK3CG(12), PIK3R1(17), PTEN(80), RACGAP1(1), RHO(1), ROCK1(3), ROCK2(4), RPS4X(1), WASF1(1) 27707675 205 139 185 32 44 40 31 32 58 0 <1.00e-15 <1.28e-14
6 TELPATHWAY Telomerase is a ribonucleotide protein that adds telomeric repeats to the 3' ends of chromosomes. AKT1, BCL2, EGFR, G22P1, HSPCA, IGF1R, KRAS2, MYC, POLR2A, PPP2CA, PRKCA, RB1, TEP1, TERF1, TERT, TNKS, TP53, XRCC5 15 AKT1(1), BCL2(3), EGFR(58), IGF1R(2), POLR2A(2), PPP2CA(1), PRKCA(3), RB1(21), TEP1(4), TERT(8), TNKS(4), TP53(89), XRCC5(1) 11328048 197 139 135 11 45 40 42 22 46 2 <1.00e-15 <1.28e-14
7 HSA04210_APOPTOSIS Genes involved in apoptosis AIFM1, AKT1, AKT2, AKT3, APAF1, ATM, BAD, BAX, BCL2, BCL2L1, BID, BIRC2, BIRC3, BIRC4, CAPN1, CAPN2, CASP10, CASP3, CASP6, CASP7, CASP8, CASP9, CFLAR, CHP, CHUK, CSF2RB, CYCS, DFFA, DFFB, ENDOG, FADD, FAS, FASLG, IKBKB, IKBKG, IL1A, IL1B, IL1R1, IL1RAP, IL3, IL3RA, IRAK1, IRAK2, IRAK3, IRAK4, MAP3K14, MYD88, NFKB1, NFKB2, NFKBIA, NGFB, NTRK1, PIK3CA, PIK3CB, PIK3CD, PIK3CG, PIK3R1, PIK3R2, PIK3R3, PIK3R5, PPP3CA, PPP3CB, PPP3CC, PPP3R1, PPP3R2, PRKACA, PRKACB, PRKACG, PRKAR1A, PRKAR1B, PRKAR2A, PRKAR2B, RELA, RIPK1, TNF, TNFRSF10A, TNFRSF10B, TNFRSF10C, TNFRSF10D, TNFRSF1A, TNFSF10, TP53, TRADD, TRAF2 79 AIFM1(4), AKT1(1), AKT2(1), AKT3(1), APAF1(1), ATM(7), BCL2(3), BID(4), BIRC2(1), BIRC3(1), CAPN1(5), CAPN2(1), CASP6(1), CHUK(4), CSF2RB(4), DFFA(1), DFFB(3), IKBKB(2), IL1R1(2), IL1RAP(2), IRAK1(2), IRAK2(3), IRAK3(2), IRAK4(2), MAP3K14(2), NFKB1(1), NFKB2(1), NFKBIA(2), NTRK1(4), PIK3CA(19), PIK3CB(2), PIK3CD(2), PIK3CG(12), PIK3R1(17), PIK3R2(5), PIK3R5(2), PPP3CB(1), PPP3R1(1), PPP3R2(1), PRKACG(1), PRKAR2A(1), PRKAR2B(1), RIPK1(1), TNFRSF10A(2), TNFRSF1A(2), TNFSF10(5), TP53(89), TRAF2(2) 32877178 234 133 199 39 59 40 36 53 44 2 <1.00e-15 <1.28e-14
8 HSA00562_INOSITOL_PHOSPHATE_METABOLISM Genes involved in inositol phosphate metabolism CARKL, FN3K, IMPA1, IMPA2, INPP1, INPP4A, INPP4B, INPP5A, INPP5B, INPP5E, INPPL1, IPMK, ISYNA1, ITGB1BP3, ITPK1, ITPKA, ITPKB, MINPP1, MIOX, OCRL, PI4KA, PI4KB, PIB5PA, PIK3C3, PIK3CA, PIK3CB, PIK3CD, PIK3CG, PIP4K2A, PIP4K2B, PIP4K2C, PIP5K1A, PIP5K1B, PIP5K1C, PIP5K3, PLCB1, PLCB2, PLCB3, PLCB4, PLCD1, PLCD3, PLCD4, PLCE1, PLCG1, PLCG2, PLCZ1, PTEN, PTPMT1, SKIP, SYNJ1, SYNJ2 47 FN3K(1), IMPA1(1), INPP1(2), INPP4A(1), INPP4B(1), INPP5A(1), INPP5B(2), INPP5E(1), INPPL1(4), ITGB1BP3(2), ITPKB(2), MINPP1(1), MIOX(1), OCRL(1), PI4KA(1), PI4KB(3), PIK3C3(1), PIK3CA(19), PIK3CB(2), PIK3CD(2), PIK3CG(12), PIP4K2A(3), PIP4K2B(1), PIP4K2C(3), PIP5K1B(2), PIP5K1C(4), PLCB1(3), PLCB2(5), PLCB3(5), PLCB4(3), PLCD1(1), PLCD3(2), PLCD4(2), PLCE1(7), PLCG1(3), PLCG2(11), PLCZ1(1), PTEN(80), SYNJ1(1), SYNJ2(7) 29194980 205 131 187 33 50 41 29 31 53 1 <1.00e-15 <1.28e-14
9 HSA04012_ERBB_SIGNALING_PATHWAY Genes involved in ErbB signaling pathway ABL1, ABL2, AKT1, AKT2, AKT3, ARAF, AREG, BAD, BRAF, BTC, CAMK2A, CAMK2B, CAMK2D, CAMK2G, CBL, CBLB, CBLC, CDKN1A, CDKN1B, CRK, CRKL, EGF, EGFR, EIF4EBP1, ELK1, ERBB2, ERBB3, ERBB4, EREG, FRAP1, GAB1, GRB2, GSK3B, HBEGF, HRAS, JUN, KRAS, MAP2K1, MAP2K2, MAP2K4, MAP2K7, MAPK1, MAPK10, MAPK3, MAPK8, MAPK9, MYC, NCK1, NCK2, NRAS, NRG1, NRG2, NRG3, NRG4, PAK1, PAK2, PAK3, PAK4, PAK6, PAK7, PIK3CA, PIK3CB, PIK3CD, PIK3CG, PIK3R1, PIK3R2, PIK3R3, PIK3R5, PLCG1, PLCG2, PRKCA, PRKCB1, PRKCG, PTK2, RAF1, RPS6KB1, RPS6KB2, SHC1, SHC2, SHC3, SHC4, SOS1, SOS2, SRC, STAT5A, STAT5B, TGFA 85 ABL1(1), ABL2(1), AKT1(1), AKT2(1), AKT3(1), ARAF(1), AREG(1), BRAF(6), CAMK2B(1), CAMK2D(1), CAMK2G(4), CBL(1), CBLB(1), CBLC(1), CDKN1A(1), CDKN1B(3), EGF(3), EGFR(58), ELK1(1), ERBB2(7), ERBB3(4), ERBB4(1), GAB1(1), GSK3B(1), KRAS(1), MAP2K1(1), MAP2K2(1), MAP2K7(1), MAPK1(3), MAPK3(2), MAPK9(2), NRAS(1), NRG1(4), NRG2(1), NRG3(1), PAK1(2), PAK3(2), PAK6(1), PAK7(3), PIK3CA(19), PIK3CB(2), PIK3CD(2), PIK3CG(12), PIK3R1(17), PIK3R2(5), PIK3R5(2), PLCG1(3), PLCG2(11), PRKCA(3), PRKCG(1), PTK2(2), RPS6KB1(1), SHC3(1), SHC4(1), SOS1(1), SOS2(4), SRC(2), STAT5A(2), STAT5B(2), TGFA(2) 40950577 223 129 181 52 58 44 49 46 26 0 <1.00e-15 <1.28e-14
10 HSA04110_CELL_CYCLE Genes involved in cell cycle ABL1, ANAPC1, ANAPC10, ANAPC11, ANAPC2, ANAPC4, ANAPC5, ANAPC7, ATM, ATR, BUB1, BUB1B, BUB3, CCNA1, CCNA2, CCNB1, CCNB2, CCNB3, CCND1, CCND2, CCND3, CCNE1, CCNE2, CCNH, CDC14A, CDC14B, CDC16, CDC2, CDC20, CDC23, CDC25A, CDC25B, CDC25C, CDC26, CDC27, CDC45L, CDC6, CDC7, CDK2, CDK4, CDK6, CDK7, CDKN1A, CDKN1B, CDKN1C, CDKN2A, CDKN2B, CDKN2C, CDKN2D, CHEK1, CHEK2, CREBBP, CUL1, DBF4, E2F1, E2F2, E2F3, EP300, ESPL1, FZR1, GADD45A, GADD45B, GADD45G, GSK3B, hCG_1982709, HDAC1, HDAC2, LOC440917, LOC728919, MAD1L1, MAD2L1, MAD2L2, MCM2, MCM3, MCM4, MCM5, MCM6, MCM7, MDM2, ORC1L, ORC2L, ORC3L, ORC4L, ORC5L, ORC6L, PCNA, PKMYT1, PLK1, PRKDC, PTTG1, PTTG2, RB1, RBL1, RBL2, RBX1, SFN, SKP1, SKP2, SMAD2, SMAD3, SMAD4, SMC1A, SMC1B, TFDP1, TGFB1, TGFB2, TGFB3, TP53, WEE1, YWHAB, YWHAE, YWHAG, YWHAH, YWHAQ, YWHAZ 109 ABL1(1), ANAPC1(4), ANAPC10(2), ANAPC11(1), ANAPC2(1), ANAPC4(2), ANAPC5(2), ATM(7), ATR(9), BUB1(4), BUB1B(1), CCNA1(1), CCNB3(1), CCND2(1), CCNE1(1), CCNE2(1), CCNH(2), CDC20(2), CDC23(1), CDC25B(1), CDC27(3), CDC7(2), CDKN1A(1), CDKN1B(3), CDKN2A(3), CDKN2B(1), CDKN2C(4), CDKN2D(1), CHEK1(5), CREBBP(8), CUL1(5), E2F2(3), E2F3(2), EP300(5), ESPL1(8), GADD45B(1), GSK3B(1), HDAC1(1), HDAC2(3), MAD1L1(1), MCM2(3), MCM3(3), MCM4(3), MCM5(1), MCM6(1), MCM7(2), MDM2(1), PCNA(1), PKMYT1(1), PLK1(1), PRKDC(5), PTTG2(2), RB1(21), RBL1(3), RBL2(1), SMAD3(1), SMAD4(1), SMC1A(7), SMC1B(3), TGFB1(1), TGFB2(1), TP53(89), YWHAB(1), YWHAE(2) 54445600 261 129 232 42 53 45 52 32 76 3 <1.00e-15 <1.28e-14

Table 7.  Get Full Table A Ranked List of Significantly Mutated Genesets (Excluding Significantly Mutated Genes). Number of significant genesets found: 0. Number of genesets displayed: 10

rank geneset description genes N_genes mut_tally N n npat nsite nsil n1 n2 n3 n4 n5 n6 p q
1 HSA04640_HEMATOPOIETIC_CELL_LINEAGE Genes involved in hematopoietic cell lineage ANPEP, CD14, CD19, CD1A, CD1B, CD1C, CD1D, CD1E, CD2, CD22, CD24, CD33, CD34, CD36, CD37, CD38, CD3D, CD3E, CD3G, CD4, CD44, CD5, CD55, CD59, CD7, CD8A, CD8B, CD9, CR1, CR2, CSF1, CSF1R, CSF2, CSF2RA, CSF3, CSF3R, DNTT, EPO, EPOR, FCER2, FCGR1A, FLT3, FLT3LG, GP1BA, GP1BB, GP5, GP9, GYPA, HLA-DRA, HLA-DRB1, HLA-DRB3, HLA-DRB4, HLA-DRB5, IL11, IL11RA, IL1A, IL1B, IL1R1, IL1R2, IL2RA, IL3, IL3RA, IL4, IL4R, IL5, IL5RA, IL6, IL6R, IL7, IL7R, IL9R, ITGA1, ITGA2, ITGA2B, ITGA3, ITGA4, ITGA5, ITGA6, ITGAM, ITGB3, KIT, KITLG, MME, MS4A1, TFRC, THPO, TNF, TPO 79 ANPEP(3), CD14(2), CD19(2), CD1B(1), CD1D(2), CD1E(1), CD22(5), CD33(3), CD34(2), CD36(2), CD38(2), CD3E(1), CD3G(1), CD44(4), CD5(2), CD55(2), CD7(2), CD8A(1), CD8B(1), CD9(2), CR1(5), CR2(6), CSF1R(3), CSF3R(2), DNTT(2), EPO(1), EPOR(1), FLT3(7), FLT3LG(1), GP9(1), GYPA(1), HLA-DRA(1), HLA-DRB1(2), HLA-DRB5(2), IL1R1(2), IL2RA(3), IL4(1), IL4R(6), IL5RA(1), IL6(2), IL6R(2), IL7(1), IL7R(1), ITGA1(1), ITGA2(4), ITGA2B(2), ITGA3(2), ITGA4(7), ITGA5(1), ITGA6(2), ITGAM(8), ITGB3(1), KIT(3), KITLG(2), MS4A1(2), TFRC(2), THPO(1), TPO(4) 31351365 137 76 135 51 54 23 28 20 12 0 0.00028 0.11
2 HSA04070_PHOSPHATIDYLINOSITOL_SIGNALING_SYSTEM Genes involved in phosphatidylinositol signaling system CALM1, CALM2, CALM3, CALML3, CALML6, CARKL, CDIPT, CDS1, CDS2, DGKA, DGKB, DGKD, DGKE, DGKG, DGKH, DGKI, DGKQ, DGKZ, FN3K, IMPA1, IMPA2, INPP1, INPP4A, INPP4B, INPP5A, INPP5B, INPP5D, INPP5E, INPPL1, ITGB1BP3, ITPK1, ITPKA, ITPKB, ITPR1, ITPR2, ITPR3, OCRL, PI4KA, PI4KB, PIB5PA, PIK3C2A, PIK3C2B, PIK3C2G, PIK3C3, PIK3CA, PIK3CB, PIK3CD, PIK3CG, PIK3R1, PIK3R2, PIK3R3, PIK3R5, PIP4K2A, PIP4K2B, PIP4K2C, PIP5K1A, PIP5K1B, PIP5K1C, PIP5K3, PLCB1, PLCB2, PLCB3, PLCB4, PLCD1, PLCD3, PLCD4, PLCE1, PLCG1, PLCG2, PLCZ1, PRKCA, PRKCB1, PRKCG, PTEN, PTPMT1, SKIP, SYNJ1, SYNJ2 69 CALM1(1), CDS1(1), DGKB(1), DGKD(6), DGKE(2), DGKG(5), DGKH(3), DGKI(3), DGKQ(1), DGKZ(5), FN3K(1), IMPA1(1), INPP1(2), INPP4A(1), INPP4B(1), INPP5A(1), INPP5B(2), INPP5D(6), INPP5E(1), INPPL1(4), ITGB1BP3(2), ITPKB(2), ITPR1(7), ITPR2(11), ITPR3(11), OCRL(1), PI4KA(1), PI4KB(3), PIK3C2A(3), PIK3C2B(2), PIK3C3(1), PIK3CB(2), PIK3CD(2), PIK3CG(12), PIK3R2(5), PIK3R5(2), PIP4K2A(3), PIP4K2B(1), PIP4K2C(3), PIP5K1B(2), PIP5K1C(4), PLCB1(3), PLCB2(5), PLCB3(5), PLCB4(3), PLCD1(1), PLCD3(2), PLCD4(2), PLCE1(7), PLCG1(3), PLCG2(11), PLCZ1(1), PRKCA(3), PRKCG(1), SYNJ1(1), SYNJ2(7) 46351323 183 92 181 48 68 29 33 25 27 1 0.00057 0.11
3 GPCRDB_CLASS_A_RHODOPSIN_LIKE ADORA1, ADORA2A, ADORA2B, ADORA3, ADRA1A, ADRA1B, ADRA1D, ADRA2A, ADRA2C, ADRB1, ADRB2, ADRB3, AGTR1, AGTR2, AGTRL1, AVPR1A, AVPR1B, AVPR2, BDKRB1, BDKRB2, BLR1, BRS3, C3AR1, C5R1, CCBP2, CCKAR, CCKBR, CCR1, CCR10, CCR2, CCR3, CCR4, CCR5, CCR6, CCR7, CCR8, CCR9, CCRL1, CCRL2, CHML, CHRM1, CHRM2, CHRM3, CHRM4, CHRM5, CMKLR1, CMKOR1, CNR1, CNR2, CX3CR1, CXCR3, CXCR4, DRD1, DRD2, DRD3, DRD4, DRD5, EDNRA, EDNRB, ELA3A, F2R, F2RL1, F2RL2, F2RL3, FPR1, FPRL1, FPRL2, FSHR, GALR1, GALR2, GALR3, GALT, GHSR, GNB2L1, GPR10, GPR147, GPR17, GPR173, GPR174, GPR23, GPR24, GPR27, GPR3, GPR30, GPR35, GPR37, GPR37L1, GPR4, GPR44, GPR50, GPR6, GPR63, GPR74, GPR77, GPR83, GPR85, GPR87, GPR92, GRPR, HCRTR1, HCRTR2, HRH1, HRH2, HRH3, HTR1A, HTR1B, HTR1D, HTR1E, HTR1F, HTR2A, HTR2B, HTR2C, HTR4, HTR5A, HTR6, HTR7, HTR7, LOC93164, IL8RA, IL8RB, LHCGR, LTB4R, MAS1, MC1R, MC3R, MC4R, MC5R, MLNR, MTNR1A, MTNR1B, NMBR, NMUR1, NMUR2, NPY1R, NPY2R, NPY5R, NPY6R, NTSR1, NTSR2, OPN1SW, OPN3, OPRD1, OPRK1, OPRL1, OPRM1, OR10A5, OR11A1, OR12D3, OR1C1, OR1F1, OR1Q1, OR2H1, OR5V1, OR5V1, OR12D3, OR7A5, OR7C1, OR8B8, OXTR, P2RY1, P2RY10, P2RY11, P2RY12, P2RY13, P2RY14, P2RY2, P2RY5, P2RY6, PPYR1, PTAFR, PTGDR, PTGER1, PTGER2, PTGER4, PTGFR, PTGIR, Rgr, RGR, RHO, RRH, SSTR1, SSTR2, SSTR3, SSTR4, SUCNR1, TBXA2R, TRHR 158 ADORA1(2), ADORA2B(1), ADORA3(1), ADRA1D(1), ADRA2C(4), ADRB1(1), ADRB2(1), AGTR1(2), AGTR2(1), AVPR1A(3), AVPR1B(1), BDKRB1(1), BDKRB2(3), BRS3(2), C3AR1(1), CCBP2(1), CCKAR(3), CCKBR(2), CCR1(2), CCR10(1), CCR3(2), CCR4(1), CCR6(2), CCR7(2), CCR8(1), CCR9(2), CHML(2), CHRM2(4), CHRM3(3), CMKLR1(1), CNR1(2), CNR2(3), CX3CR1(2), CXCR3(2), DRD1(2), DRD2(1), DRD3(1), DRD5(4), EDNRA(1), EDNRB(1), F2RL3(1), FPR1(1), FSHR(1), GALR1(3), GALR2(1), GALT(1), GHSR(4), GNB2L1(1), GPR17(1), GPR174(1), GPR27(3), GPR3(1), GPR37(3), GPR4(1), GPR50(3), GPR6(5), GPR83(4), GPR87(1), GRPR(1), HCRTR1(2), HCRTR2(1), HRH1(1), HRH2(5), HRH3(1), HTR1A(2), HTR1B(3), HTR1D(1), HTR1E(2), HTR1F(2), HTR2A(1), HTR2C(2), HTR4(1), HTR5A(4), HTR6(1), HTR7(1), LHCGR(1), MC3R(6), MC4R(3), MC5R(2), MLNR(1), NMUR1(2), NMUR2(5), NPY5R(4), NTSR1(2), OPN1SW(2), OPRK1(1), OPRL1(2), OPRM1(4), OR1F1(2), OR1Q1(1), OR2H1(4), OR7A5(1), OR7C1(1), OXTR(2), P2RY1(1), P2RY10(1), P2RY13(3), P2RY14(1), P2RY2(2), P2RY6(2), PPYR1(2), PTAFR(2), PTGDR(1), PTGER2(2), PTGER4(2), PTGFR(2), PTGIR(1), RHO(1), RRH(1), SSTR2(1), SSTR3(2), SSTR4(4), TRHR(2) 46757547 219 105 217 84 99 27 38 40 15 0 0.0007 0.11
4 HSA04060_CYTOKINE_CYTOKINE_RECEPTOR_INTERACTION Genes involved in cytokine-cytokine receptor interaction ACVR1, ACVR1B, ACVR2A, ACVR2B, AMH, AMHR2, BMP2, BMP7, BMPR1A, BMPR1B, BMPR2, CCL1, CCL11, CCL13, CCL14, CCL15, CCL16, CCL17, CCL18, CCL19, CCL2, CCL20, CCL21, CCL22, CCL23, CCL24, CCL25, CCL26, CCL27, CCL28, CCL3, CCL4, CCL5, CCL7, CCL8, CCR1, CCR2, CCR3, CCR4, CCR5, CCR6, CCR7, CCR8, CCR9, CD27, CD40, CD40LG, CD70, CLCF1, CNTF, CNTFR, CRLF2, CSF1, CSF1R, CSF2, CSF2RA, CSF2RB, CSF3, CSF3R, CTF1, CX3CL1, CX3CR1, CXCL1, CXCL10, CXCL11, CXCL12, CXCL13, CXCL14, CXCL16, CXCL2, CXCL3, CXCL5, CXCL6, CXCL9, CXCR3, CXCR4, CXCR6, EDA, EDA2R, EDAR, EGF, EGFR, EPO, EPOR, FAS, FASLG, FLJ78302, FLT1, FLT3, FLT3LG, FLT4, GDF5, GH1, GH2, GHR, HGF, IFNA1, IFNA10, IFNA13, IFNA14, IFNA16, IFNA17, IFNA2, IFNA21, IFNA4, IFNA5, IFNA6, IFNA7, IFNA8, IFNAR1, IFNAR2, IFNB1, IFNE1, IFNG, IFNGR1, IFNGR2, IFNK, IFNW1, IL10, IL10RA, IL10RB, IL11, IL11RA, IL12A, IL12B, IL12RB1, IL12RB2, IL13, IL13RA1, IL15, IL15RA, IL17A, IL17B, IL17RA, IL17RB, IL18, IL18R1, IL18RAP, IL19, IL1A, IL1B, IL1R1, IL1R2, IL1RAP, IL2, IL20, IL20RA, IL21, IL21R, IL22, IL22RA1, IL22RA2, IL23A, IL23R, IL24, IL25, IL26, IL28A, IL28B, IL28RA, IL29, IL2RA, IL2RB, IL2RG, IL3, IL3RA, IL4, IL4R, IL5, IL5RA, IL6, IL6R, IL6ST, IL7, IL7R, IL8, IL8RA, IL8RB, IL9, IL9R, INHBA, INHBB, INHBC, INHBE, KDR, KIT, KITLG, LEP, LEPR, LIF, LIFR, LOC728045, LTA, LTB, LTBR, MET, MPL, NGFR, OSM, OSMR, PDGFB, PDGFC, PDGFRA, PDGFRB, PF4, PF4V1, PLEKHO2, PPBP, PRL, PRLR, RELT, TGFB1, TGFB2, TGFB3, TGFBR1, TGFBR2, TNF, TNFRSF10A, TNFRSF10B, TNFRSF10C, TNFRSF10D, TNFRSF11A, TNFRSF11B, TNFRSF12A, TNFRSF13B, TNFRSF13C, TNFRSF14, TNFRSF17, TNFRSF18, TNFRSF19, TNFRSF1A, TNFRSF1B, TNFRSF21, TNFRSF25, TNFRSF4, TNFRSF6B, TNFRSF8, TNFRSF9, TNFSF10, TNFSF11, TNFSF12, TNFSF13, TNFSF13B, TNFSF14, TNFSF15, TNFSF18, TNFSF4, TNFSF8, TNFSF9, TPO, TSLP, VEGFA, VEGFB, VEGFC, XCL1, XCL2, XCR1 238 ACVR1(3), ACVR2B(1), AMHR2(3), BMPR1A(2), BMPR1B(1), BMPR2(2), CCL13(1), CCL15(1), CCL19(1), CCL2(2), CCL20(1), CCL22(1), CCL24(1), CCL26(1), CCL3(1), CCL5(1), CCL7(1), CCR1(2), CCR3(2), CCR4(1), CCR6(2), CCR7(2), CCR8(1), CCR9(2), CD70(1), CLCF1(1), CNTF(1), CSF1R(3), CSF2RB(4), CSF3R(2), CX3CL1(1), CX3CR1(2), CXCL1(1), CXCL13(1), CXCL6(3), CXCR3(2), CXCR6(1), EDA(3), EGF(3), EPO(1), EPOR(1), FLT1(2), FLT3(7), FLT3LG(1), FLT4(3), GDF5(2), GH1(1), GH2(2), GHR(3), HGF(4), IFNA13(1), IFNA14(1), IFNA17(1), IFNA21(1), IFNA5(1), IFNA7(2), IFNAR2(1), IFNB1(1), IFNG(1), IFNGR1(1), IFNGR2(1), IFNW1(1), IL10RA(1), IL12B(1), IL12RB1(4), IL12RB2(4), IL13RA1(1), IL15RA(1), IL17RB(1), IL18R1(1), IL18RAP(4), IL1R1(2), IL1RAP(2), IL2(1), IL21(1), IL21R(4), IL22(3), IL22RA1(1), IL23R(2), IL24(1), IL26(1), IL28A(1), IL28B(1), IL28RA(1), IL2RA(3), IL2RB(1), IL4(1), IL4R(6), IL5RA(1), IL6(2), IL6R(2), IL7(1), IL7R(1), INHBA(3), INHBB(2), INHBE(1), KDR(10), KIT(3), KITLG(2), LEPR(1), LIFR(1), LTBR(1), MET(5), MPL(2), NGFR(1), OSM(1), OSMR(2), PDGFC(3), PDGFRB(5), PPBP(2), PRL(3), RELT(1), TGFB1(1), TGFB2(1), TGFBR1(1), TGFBR2(5), TNFRSF10A(2), TNFRSF11A(3), TNFRSF13B(2), TNFRSF14(1), TNFRSF18(1), TNFRSF1A(2), TNFRSF1B(2), TNFRSF25(1), TNFRSF6B(2), TNFRSF8(3), TNFSF10(5), TNFSF11(4), TNFSF13B(2), TNFSF14(3), TNFSF18(1), TNFSF4(1), TNFSF8(1), TNFSF9(3), TPO(4), VEGFA(2), VEGFC(2), XCL1(2), XCL2(2), XCR1(1) 66065757 271 115 270 80 94 38 60 45 34 0 0.00075 0.11
5 HSA00040_PENTOSE_AND_GLUCURONATE_INTERCONVERSIONS Genes involved in pentose and glucuronate interconversions AKR1B1, DCXR, GUSB, RPE, UGDH, UGP2, UGT1A1, UGT1A10, UGT1A3, UGT1A4, UGT1A5, UGT1A6, UGT1A7, UGT1A8, UGT1A9, UGT2A1, UGT2A3, UGT2B10, UGT2B11, UGT2B15, UGT2B17, UGT2B28, UGT2B4, UGT2B7, XYLB 24 GUSB(3), RPE(1), UGP2(1), UGT1A1(1), UGT1A10(1), UGT1A3(1), UGT1A5(2), UGT1A6(2), UGT1A9(1), UGT2A1(5), UGT2A3(3), UGT2B10(1), UGT2B11(3), UGT2B17(1), UGT2B28(5), UGT2B4(5), UGT2B7(3), XYLB(4) 9832004 43 32 42 16 12 6 11 10 4 0 0.001 0.12
6 FXRPATHWAY The nuclear receptor transcription factors FXR and LXR are activated by cholesterol metabolites and regulate cholesterol homeostasis. FABP6, LDLR, NR0B2, NR1H3, NR1H4, RXRA 6 LDLR(4), NR0B2(3), NR1H3(1), NR1H4(3), RXRA(3) 2161711 14 14 14 3 3 2 4 3 2 0 0.0011 0.12
7 HSA04610_COMPLEMENT_AND_COAGULATION_CASCADES Genes involved in complement and coagulation cascades A2M, BDKRB1, BDKRB2, C1QA, C1QB, C1QC, C1R, C1S, C2, C3, C3AR1, C4A, C4B, C4BPA, C4BPB, C5, C5AR1, C6, C7, C8A, C8B, C8G, C9, CD46, CD55, CD59, CFB, CFD, CFH, CFI, CPB2, CR1, CR2, F10, F11, F12, F13A1, F13B, F2, F2R, F3, F5, F7, F8, F9, FGA, FGB, FGG, KLKB1, KNG1, MASP1, MASP2, MBL2, PLAT, PLAU, PLAUR, PLG, PROC, PROS1, SERPINA1, SERPINA5, SERPINC1, SERPIND1, SERPINE1, SERPINF2, SERPING1, TFPI, THBD, VWF 67 A2M(5), BDKRB1(1), BDKRB2(3), C1QB(1), C1QC(1), C1R(1), C1S(1), C2(2), C3(8), C3AR1(1), C5(1), C5AR1(1), C6(3), C7(6), C8A(3), C8B(5), C9(1), CD55(2), CFB(1), CFH(3), CFI(1), CPB2(1), CR1(5), CR2(6), F10(1), F13A1(5), F13B(2), F2(1), F5(9), F7(1), F8(8), F9(4), FGA(3), FGB(1), FGG(4), KLKB1(1), KNG1(4), MASP1(2), MBL2(1), PLAU(2), PLAUR(1), PLG(2), PROC(2), SERPINA1(3), SERPINA5(1), SERPINC1(2), SERPIND1(1), SERPINE1(1), SERPINF2(1), SERPING1(5), THBD(3), VWF(10) 35749445 144 75 141 68 48 23 33 19 20 1 0.0021 0.18
8 CTLPATHWAY Cytotoxic T lymphocytes induce apoptosis in infected cells presenting antigen-MHC-I complexes via the perforin and Fas/Fas ligand pathways. B2M, CD3D, CD3E, CD3G, CD3Z, GZMB, HLA-A, ICAM1, ITGAL, ITGB2, PRF1, TNFRSF6, TNFSF6, TRA@, TRB@ 10 B2M(2), CD3E(1), CD3G(1), GZMB(3), HLA-A(1), ITGAL(4), ITGB2(5), PRF1(3) 3417489 20 15 20 14 6 1 4 1 8 0 0.0047 0.35
9 BLOOD_CLOTTING_CASCADE F10, F11, F12, F13B, F2, F5, F7, F8, F8A1, F9, FGA, FGB, FGG, LPA, PLG, PLAT, PLAU, PLG, SERPINB2, SERPINE1, SERPINF2, VWF 20 F10(1), F13B(2), F2(1), F5(9), F7(1), F8(8), F9(4), FGA(3), FGB(1), FGG(4), LPA(6), PLAU(2), PLG(2), SERPINB2(2), SERPINE1(1), SERPINF2(1), VWF(10) 14027949 58 43 58 22 21 10 11 5 11 0 0.0052 0.35
10 HSA00980_METABOLISM_OF_XENOBIOTICS_BY_CYTOCHROME_P450 Genes involved in metabolism of xenobiotics by cytochrome P450 ADH1A, ADH1B, ADH1C, ADH4, ADH5, ADH6, ADH7, ADHFE1, AKR1C1, AKR1C2, AKR1C3, AKR1C4, ALDH1A3, ALDH3A1, ALDH3B1, ALDH3B2, CYP1A1, CYP1A2, CYP1B1, CYP2B6, CYP2C18, CYP2C19, CYP2C8, CYP2C9, CYP2E1, CYP2F1, CYP2S1, CYP3A4, CYP3A43, CYP3A5, CYP3A7, DHDH, EPHX1, GSTA1, GSTA2, GSTA3, GSTA4, GSTA5, GSTK1, GSTM1, GSTM2, GSTM3, GSTM4, GSTM5, GSTO2, GSTP1, GSTT1, GSTT2, GSTZ1, MGST1, MGST2, MGST3, UGT1A1, UGT1A10, UGT1A3, UGT1A4, UGT1A5, UGT1A6, UGT1A7, UGT1A8, UGT1A9, UGT2A1, UGT2A3, UGT2B10, UGT2B11, UGT2B15, UGT2B17, UGT2B28, UGT2B4, UGT2B7 68 ADH1A(4), ADH1B(3), ADH1C(3), ADH6(2), ADH7(1), AKR1C1(3), AKR1C4(2), ALDH1A3(1), ALDH3A1(3), ALDH3B1(2), ALDH3B2(1), CYP1A1(1), CYP1B1(4), CYP2B6(2), CYP2C18(1), CYP2C8(2), CYP2C9(2), CYP2E1(4), CYP2S1(1), CYP3A4(3), CYP3A43(2), CYP3A7(3), DHDH(1), GSTA1(2), GSTA4(2), GSTA5(1), GSTK1(2), GSTM1(1), GSTM3(1), GSTP1(1), GSTT1(1), MGST1(1), MGST3(1), UGT1A1(1), UGT1A10(1), UGT1A3(1), UGT1A5(2), UGT1A6(2), UGT1A9(1), UGT2A1(5), UGT2A3(3), UGT2B10(1), UGT2B11(3), UGT2B17(1), UGT2B28(5), UGT2B4(5), UGT2B7(3) 21984496 98 57 97 32 38 8 24 25 3 0 0.0063 0.37
Methods & Data
Methods

In brief, we tabulate the number of mutations and the number of covered bases for each gene. The counts are broken down by mutation context category: four context categories that are discovered by MutSig, and one for indel and 'null' mutations, which include indels, nonsense mutations, splice-site mutations, and non-stop (read-through) mutations. For each gene, we calculate the probability of seeing the observed constellation of mutations, i.e. the product P1 x P2 x ... x Pm, or a more extreme one, given the background mutation rates calculated across the dataset. [1]

Download Results

This is an experimental feature. The full results of the analysis summarized in this report can be downloaded from the TCGA Data Coordination Center.

References
[1] TCGA, Integrated genomic analyses of ovarian carcinoma, Nature 474:609 Ð 615 (2011)