A Comparison of Computational Tools for Prediction of Cancer Driver Genes

Abstract

The significant improvements in throughput and quality of DNA sequencing technology have revolutionized our ability to identify the genetic sequence of human cells. High-throughput genome sequencing of tumor cells has furthermore enabled us to identify the complete spectrum of somatically acquired mutations of individual tumors. Among the thousands of somatic mutations that can be found in a given tumor, only a limited number are likely to be of importance for cancer development. A central challenge in cancer genomics research is thus to identify the mutations that are causally implicated in tumorigenesis, commonly known as cancer driver mutations. Genes that carry driver mutations are known as cancer driver genes. Large-scale bioinformatics analysis of tumor genomes have exploited different strategies in order to distinguish positively selected driver mutations from their neutral counterparts. The different computational approaches are frequently implemented as stand-alone software tools, allowing individual researchers with tumor sequencing data to predict likely cancer driver genes. The actual installation and application of bioinformatics tools can be cumbersome for cancer researchers with limited computational competence, and the comparative performance of driver gene prediction results with different approaches and algorithms would therefore be difficult to obtain. To this end, we have implemented a single computational workflow for driver gene prediction within the Galaxy framework, a user-friendly web-based platform for data intensive biomedical research. Our workflow accepts a single input file with tumor DNA variation data and will subsequently run three of the most commonly used tools for cancer driver prediction, that is, IntoGen, MutSigCV, and DrGap. A report is generated that indicates the comparative performance of the individual tools (i.e. where the tools are in agreement, and where they are not) as well as simple visualization of their overlapping predictions.

The significant improvements in throughput and quality of DNA sequencing technology have revolutionized our ability to identify the genetic sequence of human cells. High-throughput genome sequencing of tumor cells has furthermore enabled us to identify the complete spectrum of somatically acquired mutations of individual tumors. Among the thousands of somatic mutations that can be found in a given tumor, only a limited number are likely to be of importance for cancer development. A central challenge in cancer genomics research is thus to identify the mutations that are causally implicated in tumorigenesis, commonly known as cancer driver mutations. Genes that carry driver mutations are known as cancer driver genes. Large-scale bioinformatics analysis of tumor genomes have exploited different strategies in order to distinguish positively selected driver mutations from their neutral counterparts. The different computational approaches are frequently implemented as stand-alone software tools, allowing individual researchers with tumor sequencing data to predict likely cancer driver genes. The actual installation and application of bioinformatics tools can be cumbersome for cancer researchers with limited computational competence, and the comparative performance of driver gene prediction results with different approaches and algorithms would therefore be difficult to obtain. To this end, we have implemented a single computational workflow for driver gene prediction within the Galaxy framework, a user-friendly web-based platform for data intensive biomedical research. Our workflow accepts a single input file with tumor DNA variation data and will subsequently run three of the most commonly used tools for cancer driver prediction, that is, IntoGen, MutSigCV, and DrGap. A report is generated that indicates the comparative performance of the individual tools (i.e. where the tools are in agreement, and where they are not) as well as simple visualization of their overlapping predictions.