Poster #33: Naeem Ahmed

Optimizing copy number profiles for tumor cell phylogeny inference from low-coverage single-cell sequencing data

Naeem Ahmed, BS, School of Computing, University of Connecticut, Storrs, Connecticut, USA. Samson Weiner, PhD, School of Computing, University of Connecticut, Storrs, Connecticut, USA. Carter Payne, BS, School of Computing, University of Connecticut, Storrs, Connecticut, USA. Mukul S. Bansal, PhD, School of Computing, University of Connecticut, Storrs, Connecticut, USA; The Institute for Systems Genomics, University of Connecticut, Storrs, Connecticut, USA.

Accurate reconstruction of tumor cell phylogenies is crucial for understanding cancer evolutionary dynamics, inferring intra-tumor heterogeneity, and designing effective and personalized cancer treatment strategies. One of the most promising and cost-effective approaches for reconstructing tumor phylogenies for large numbers of cells is to use copy number alteration (CNA) data obtained from low-coverage single-cell DNA sequencing of sampled tumor cells. However, errors and uncertainty in inferred copy number profiles (CNPs) due to low sequence coverage can have a drastic negative effect on the accuracy of tumor cell phylogeny reconstruction. In this work, we show that careful selection of bin size, a critical tuning parameter for inferring CNPs, is necessary to maximize the accuracy of tumor cell phylogeny inference. We first use three previously published low-coverage single-cell DNA sequencing (scDNAseq) datasets from real tumors to demonstrate that even small changes in bin size can substantially impact downstream tumor phylogeny reconstruction. Then, we show that this effect persists even when using simulated low-coverage datasets generated under idealized conditions with minimal noise. Finally, we propose a simple computational approach (called bin-selector) to select bin sizes that maximize tumor phylogeny accuracy. Our experiments span a wide range of bin sizes and account for the impact of various factors, including coverage level, phylogeny reconstruction method, and CNP inference method, explicitly accounting for phylogenetic reconstruction uncertainty. Our evaluation of the proposed bin-selector shows that it can substantially improve the accuracy of tumor cell phylogeny reconstruction compared to baseline methods. Most importantly, these findings establish bin size as a major parameter for accurate reconstruction of downstream phylogenies and demonstrate the importance of data-driven bin size selection for low-coverage scDNAseq data.