A deep learning system accurately classifies primary and metastatic cancers using passenger mutation patterns
Wei Jiao, Gurnit Atwal, Paz Polak, Rosa Karlić, Edwin Cuppen, Fátima Al‐Shahrour, Gurnit Atwal, Peter J. Bailey, Andrew V. Biankin, Paul C. Boutros, Peter J. Campbell, David K. Chang, Susanna L. Cooke, Vikram Deshpande, Bishoy M. Faltas, William C. Faquin, Levi A. Garraway, Gad Getz, Sean M. Grimmond, Syed Haider, Katherine A. Hoadley, Wei Jiao, Vera B. Kaiser, Rosa Karlić, Mamoru Kato, Kirsten Kübler, Alexander J. Lazar, Constance H. Li, David N. Louis, Adam Margolin, Sancha Martin, Hardeep K. Nahal-Bose, G. Petur Nielsen, Serena Nik‐Zainal, Larsson Omberg, Christine P’ng, Marc D. Perry, Paz Polak, Esther Rheinbay, Mark A. Rubin, Colin A. Semple, Dennis C. Sgroi, Tatsuhiro Shibata, Reiner Siebert, John A. Smith, Lincoln Stein, Miranda D. Stobbe, Ren Sun, Kevin Thai, Derek Wright, Chin‐Lee Wu, Ke Yuan, Junjun Zhang, Alexandra Danyi, Jeroen de Ridder, Carla van Herpen, Martijn P. Lolkema, Neeltje Steeghs, Gad Getz, Quaid Morris, Lincoln Stein, Lauri A. Aaltonen, Federico Abascal, Adam Abeshouse, Hiroyuki Aburatani, David J. Adams, Nishant Agrawal, Keun Soo Ahn, Sung-Min Ahn, Hiroshi Aikata, Rehan Akbani, Kadir C. Akdemir, Hikmat Al‐Ahmadie, Sultan T. Al‐Sedairy, Fátima Al‐Shahrour, Malik Alawi, Monique Albert, Kenneth Aldape, Ludmil B. Alexandrov, Adrian Ally, Kathryn Alsop, Eva G. Álvarez, Fernanda Amary, Samirkumar B. Amin, Brice Aminou, Ole Ammerpohl, Matthew J. Anderson, Yeng Ang, Davide Antonello, Pavana Anur, Samuel Aparício, Elizabeth L. Appelbaum, Yasuhito Arai, Axel Aretz, Koji Arihiro, Shun‐ichi Ariizumi, Joshua Armenia, Laurent Arnould, L. Sylvia, Yassen Assenov
Abstract
In cancer, the primary tumour's organ of origin and histopathology are the strongest determinants of its clinical behaviour, but in 3% of cases a patient presents with a metastatic tumour and no obvious primary. Here, as part of the ICGC/TCGA Pan-Cancer Analysis of Whole Genomes (PCAWG) Consortium, we train a deep learning classifier to predict cancer type based on patterns of somatic passenger mutations detected in whole genome sequencing (WGS) of 2606 tumours representing 24 common cancer types produced by the PCAWG Consortium. Our classifier achieves an accuracy of 91% on held-out tumor samples and 88% and 83% respectively on independent primary and metastatic samples, roughly double the accuracy of trained pathologists when presented with a metastatic tumour without knowledge of the primary. Surprisingly, adding information on driver mutations reduced accuracy. Our results have clinical applicability, underscore how patterns of somatic passenger mutations encode the state of the cell of origin, and can inform future strategies to detect the source of circulating tumour DNA.