Combining tumor genome simulation with crowdsourcing to benchmark somatic single-nucleotide-variant detection

Area of Science:

• Genomics
• Bioinformatics
• Cancer Research

Background:

• Somatic mutation detection is crucial for understanding cancer progression, survival, and therapy response.
• Benchmarking mutation detection tools is challenging due to the lack of gold standards and data sharing issues.

Purpose of the Study:

• To establish a crowdsourced benchmark for somatic mutation detection algorithms.
• To introduce BAMSurgeon, a tool for simulating cancer genomes for benchmarking.

Main Methods:

• The ICGC-TCGA DREAM Somatic Mutation Calling Challenge was launched.
• BAMSurgeon was used to create three in silico tumors for analysis.
• 248 analyses were performed using various somatic mutation detection algorithms.

Main Results:

• Different algorithms demonstrated characteristic error profiles.
• False positive mutations in simulated tumors mirrored trinucleotide profiles found in human tumors.
• Ensemble pipelines consistently outperformed individual pipelines across all simulations.

Conclusions:

• BAMSurgeon facilitates the creation of simulated cancer genomes for robust algorithm benchmarking.
• Ensemble approaches enhance the accuracy of somatic mutation detection.
• This work addresses critical needs in cancer genomics tool assessment and development.