Poster Presentation & Flash Talk Presentation 29th Lorne Cancer Conference 2017

SVclone: inference of cancer cell fraction using structural variation (#15)

Marek Cmero 1 2 , Ke Yuan 3 , Kangbo Mo 4 , Niall M Corcoran 5 , Tony Papenfuss 2 , Florian Markowetz 6 , Christopher M Hovens 5 , Cheng Soon-Ong 7 8 , Geoff Macintyre 6 9
  1. Surgery, University of Melbourne, Parkville, VIC, Australia
  2. Bioinformatics, Walter + Eliza Hall Institute, Melbourne, VIC, Australia
  3. University of Glasgow, Glasgow, Scotland, UK
  4. Computing & Information Systems, University of Melbourne, Melbourne, VIC, Australia
  5. Department of Surgery, Royal Melbourne Hospital and the Australian Prostate Cancer Research Centre Epworth, Richmond, VIC
  6. Cancer Research UK Cambridge Institute, Cambridge
  7. Computer Science, Australian National University, Canberra
  8. Machine Learning Research Group, Data61, Canberra, ACT, Australia
  9. Computing and Information Systems, University of Melbourne, Melbourne


Understanding intra-tumour heterogeneity has fundamental implications for the treatment and prognostication of cancer. Many approaches have arisen for inferring the evolutionary dynamics and clonality landscapes of tumour cell populations from point-mutation and copy-number data. However, there are currently no methods that fully incorporate structural variation (SV). Copy-number neutral rearrangements are common in certain cancers such as prostate; however, there are currently no methods to investigate the clonality status of rearrangements. To address this gap, we present SVclone, the first integrated software package for investigating the clonality of tumour samples using SV calls from modest-coverage whole-genome sequencing data.



We have developed a method which extracts and counts read information from whole-genome alignments, given a list of SV calls. We take these counts and infer variant allele frequencies (VAFs), which we input into a Bayesian model, which employs a cluster clustering process to dynamically infer the number of tumour populations and their relative proportions. We also incorporate copy-number states to correct allele frequencies in order to obtain a robust estimate of cancer cell fractions.



We tested our approach using simulations, and developed a series of corrections to account for VAF biases. In order to test the algorithm on real, noisy tumour data, we created in-silico mixtures of prostate cancer metastases samples in known proportions and were able to recapitulate the true proportions and cluster numbers with high accuracy using a fraction of the data points that SNV-based methods use. We then applied our algorithm to several different cancer samples, uncovering genotypic differences in SNV and SV-driven clones, as well as discovering examples where inferring cancer cell fraction from SVs is advantageous to SNV-derived methods.



The evolution of tumours driven by large-scale (particularly balanced) rearrangements, such as prostate cancer, has not been fully explored. We have developed a tool to infer the clonality of whole-genome sequenced tumour samples, giving accurate results with a relatively small number of data points. This method shows great potential for obtaining insight into SV evolution in the context of clonal architecture.