Crowd-sourced benchmarking of single-sample tumour subclonal reconstruction

• DOI: 10.1101/2022.06.14.495937
• Publication Date: 2022-06-15T19:00:10Z
• Authors: Adriana Salcedo; Maxime Tarabichi; Alex Buchanan; Shadrielle M.G. E...; Hongjiu Zhang; Kaiyi Zhu; Tai-Hsien Ou Yang; Ignaty Leshchiner; Dimitris Anastassiou; Yuanfang Guan; Gun Ho Jang; Kerstin Haase; Amit G. Deshwar; William Zou; Imaad Umar; Stefan Dentro; Jeff A. Wintersinger; Kami Chiotti; Jonas Demeulemeester; Clemency Jolly; Lesia Sycza; Minjeong Ko; ; ; David C. Wedge; Quaid D. Morris; Kyle Ellrott; Peter Van Loo; Paul C. Boutros
• Keywords: Benchmark (surveying); Benchmarking; Sample (material); Code (set theory)

Abstract Tumours are dynamically evolving populations of cells. Subclonal reconstruction algorithms use bulk DNA sequencing data to quantify parameters tumour evolution, allowing assessment how cancers initiate, progress and respond selective pressures. A plethora subclonal have been created, but their relative performance across the varying biological technical features real-world cancer genomic is unclear. We therefore launched ICGC-TCGA DREAM Somatic Mutation Calling -- Tumour Heterogeneity Evolution Challenge. This seven-year community effort used cloud-computing benchmark 31 containerized on 51 simulated tumours. Each algorithm was scored for accuracy seven independent tasks, leading 12,061 total runs. Algorithm choice influenced significantly more than features, purity-adjusted read-depth, copy number state read mappability were associated with most tasks. No single a top performer all tasks existing ensemble strategies surprisingly unable outperform best individual methods, highlighting key research need. All evaluation code datasets available support further determinants development improved methods understand evolution.