mSignatureDB: a database for deciphering mutational signatures in human cancers

Abstract

Cancer is a genetic disease caused by somatic mutations; however, the understanding of the causative biological processes generating these mutations is limited. A cancer genome bears the cumulative effects of mutational processes during tumor development. Deciphering mutational signatures in cancer is a new topic in cancer research. The Wellcome Trust Sanger Institute (WTSI) has categorized 30 reference signatures in the COSMIC database based on the analyses of ∼10 000 sequencing datasets from TCGA and ICGC. Large cohorts and bioinformatics skills are required to perform the same analysis as WTSI. The quantification of known signatures in custom cohorts is not possible under the current framework of the COSMIC database, which motivates us to construct a database for mutational signatures in cancers and make such analyses more accessible to general researchers. mSignatureDB (http://tardis.cgu.edu.tw/msignaturedb) integrates R packages and in-house scripts to determine the contributions of the published signatures in 15 780 individual tumors from 73 TCGA/ICGC cancer projects, making comparison of signature patterns within and between projects become possible. mSignatureDB also allows users to perform signature analysis on their own datasets, quantifying contributions of signatures at sample resolution, which is a unique feature of mSignatureDB not available in other related databases.

Genomic variants concurrently listed in a somatic and a germline mutation database have implications for disease-variant discovery and genomic privacy

“…We tested whether the prevalence of the various mutational signatures per somatic sample could predict the cSNV rate by somatic sample. These prevalences were obtained from Huang et al [26] The most significant correlation was that the prevalence of Signature 1 by somatic sample predicts 31% of the variation of cSNV rates by somatic sample (p value < 2e-16) ( Figure 3), which reinforces recent findings that the endogenous mutation process Signature 1 is active in both the soma and germline. Similarly, the median prevalence of Signature 1 by cancer type was strongly correlated with the median cSNV rate by cancer type (Figure 4).…”

Higher prevalence of homologous recombination-deficiency in lung squamous carcinoma from African Americans

“…Mutation signature 3 was mined from mSignatureDB 42 , a database of mutation signatures for more than 15,000 tumor samples from more than 73 projects.…”

Computational tools to detect signatures of mutational processes in DNA from tumours: a review and empirical comparison of performance

“…A current practice consists in first performing a de novo extraction of signatures followed by a comparison of the newly identified signatures with the reference signatures (e.g. COSMIC signatures) by means of a similarity score, typically cosine similarity ranging from 0 (completely different) to 1 (identical) [3,6]. Finally, a "novel" signature is considered to reflect a specific reference signature if the similarity is larger than a fixed cut-off.…”

“…Signatures identified with de novo methods can be compared to reference signatures (i.e. those listed in COSMIC) through measures such as cosine [5] or bootstrapped cosine similarity [6], which is a distance metric between two non-zeros vectors. The issue around the results from these comparisons is the choice of a cut-off, which implies the level at which two signatures will be considered as similar or not [3,7] Recently, [8] published a paper that has been limited to the summary of current methodologies, in particular the mathematical models and computational techniques, which form the basis of mutational signature analysis.…”