RESOURCE https://doi.org/10.1038/s41587-019-0364-z 1 Ontario Institute for Cancer Research, Toronto, Canada. 2 Department of Medical Biophysics, University of Toronto, Toronto, Canada. 3 The Francis Crick Institute, London, UK. 4 Wellcome Trust Sanger Institute, Hinxton, UK. 5 The Edward S. Rogers Senior Department of Electrical & Computer Engineering, Toronto, Canada. 6 Donnelly Centre, University of Toronto, Toronto, Canada. 7 Department of Computational Medicine and Bioinformatics, University of Michigan, Ann Arbor, MI, USA. 8 Department of Systems Biology, Columbia University, New York, NY, USA. 9 Center for Cancer Systems Therapeutics, Columbia University, New York, NY, USA. 10 Department of Electrical Engineering, Columbia University, New York, NY, USA. 11 Department of Biochemistry and Molecular Biology, The Pennsylvania State University, University Park, PA, USA. 12 Oregon Health & Sciences University, Portland, OR, USA. 13 A full list of authors and affiliations appears at the end of the paper. 14 Department of Biomolecular Engineering, Center for Biomolecular Sciences and Engineering, University of California, Santa Cruz, Santa Cruz, CA, USA. 15 Herbert Irving Comprehensive Cancer Center, Columbia University, New York, USA. 16 Department of Internal Medicine, University of Michigan, Ann Arbor, MI, USA. 17 Department of Electronic Engineering and Computer Science, University of Michigan, Ann Arbor, MI, USA. 18 Mater Research Institute, University of Queensland, Woolloongabba, Queensland, Australia. 19 Big Data Institute, University of Oxford, Oxford, UK. 20 Oxford NIHR Biomedical Research Centre, Oxford, UK. 21 Computational and Systems Biology, Memorial Sloan Kettering Cancer Center, New York, NY, USA. 22 Vector Institute for Artificial Intelligence, Toronto, Canada. 23 Department of Human Genetics, University of Leuven, Leuven, Belgium. 24 Department of Pharmacology and Toxicology, University of Toronto, Toronto, Canada. 25 Department of Human Genetics, University of California, Los Angeles, Los Angeles, CA, USA. 26 Department of Urology, University of California, Los Angeles, Los Angeles, CA, USA. 27 Institute for Precision Health, University of California, Los Angeles, Los Angeles, CA, USA. 28 Jonsson Comprehensive Cancer Center, University of California, Los Angeles, Los Angeles, CA, USA. 29 These authors contributed equally: Adriana Salcedo, Maxime Tarabichi, Shadrielle Melijah G. Espiritu, Amit G. Deshwar. 30 These authors jointly supervised this work: Kyle Ellrott, David C. Wedge, Quaid Morris, Peter Van Loo, Paul C. Boutros. *e-mail: pboutros@mednet.ucla.edu M ost tumors arise from a single ancestral cell, whose genome acquires one or more somatic driver mutations 1,2 , which give it a fitness advantage over its neighbors by manifesting hallmark characteristics of cancers 3 . This ancestral cell and its descendants proliferate, ultimately giving rise to all cancer- ous cells within the tumor. Over time, they accumulate mutations, some leading to further fitness advantages. Eventually local clonal expansions can create subpopulations of tumor cells sharing sub- sets of mutations, termed subclones. As the tumor extends spatially beyond its initial location, spatial variability can arise as different regions harbor independently evolving tumor cells with distinctive genetic and nongenetic characteristics 4–9 . DNA sequencing of tumors allows quantification of the fre- quency of specific mutations based on measurements of the fraction of mutant sequencing reads, the copy number state of the locus and the tumor purity 10,11 . By aggregating these noisy frequency measure- ments across mutations, a tumor sample’s subclonal architecture can be reconstructed from bulk sequencing data 6,11 . Subclonal recon- struction methods have proliferated rapidly in recent years 12–15 , and have revealed key characteristics of tumor evolution 4,7,16–20 , spread 21–23 and response to therapy 24,25 . Nevertheless, there has been no rigorous benchmarking of the relative or absolute accuracy of approaches for subclonal reconstruction. There are several reasons why such benchmarking has not yet been performed. First, it is difficult to identify a gold-standard truth for subclonal reconstruction. While single-cell sequencing could provide ground truth, it has pervasive errors 26 , and existing DNA-based datasets do not have sufficient depth and breadth to A community effort to create standards for evaluating tumor subclonal reconstruction Adriana Salcedo 1,2,29 , Maxime Tarabichi 3,4,29 , Shadrielle Melijah G. Espiritu 1,29 , Amit G. Deshwar 5,29 , Matei David 1 , Nathan M. Wilson 1 , Stefan Dentro 3,4 , Jeff A. Wintersinger 6 , Lydia Y. Liu 1 , Minjeong Ko 1 , Srinivasan Sivanandan 1 , Hongjiu Zhang 7 , Kaiyi Zhu  8,9,10 , Tai-Hsien Ou Yang  8,9,10 , John M. Chilton  11 , Alex Buchanan 12 , Christopher M. Lalansingh 1 , Christine P’ng 1 , Catalina V. Anghel 1 , Imaad Umar 1 , Bryan Lo 1 , William Zou 1 , DREAM SMC-Het Participants 13 , Jared T. Simpson 1 , Joshua M. Stuart  14 , Dimitris Anastassiou 8,9,10,15 , Yuanfang Guan 7,16,17 , Adam D. Ewing 18 , Kyle Ellrott  11,12,30 , David C. Wedge  19,20,30 , Quaid Morris 1,6,21,22,30 , Peter Van Loo  3,23,30 and Paul C. Boutros  2,24,25,26,27,28,30 * Tumor DNA sequencing data can be interpreted by computational methods that analyze genomic heterogeneity to infer evolu- tionary dynamics. A growing number of studies have used these approaches to link cancer evolution with clinical progression and response to therapy. Although the inference of tumor phylogenies is rapidly becoming standard practice in cancer genome analyses, standards for evaluating them are lacking. To address this need, we systematically assess methods for reconstructing tumor subclonality. First, we elucidate the main algorithmic problems in subclonal reconstruction and develop quantitative met- rics for evaluating them. Then we simulate realistic tumor genomes that harbor all known clonal and subclonal mutation types and processes. Finally, we benchmark 580 tumor reconstructions, varying tumor read depth, tumor type and somatic variant detection. Our analysis provides a baseline for the establishment of gold-standard methods to analyze tumor heterogeneity. NATURE BIOTECHNOLOGY | VOL 38 | JANUARY 2020 | 97–107 | www.nature.com/naturebiotechnology 97