DICE: Fast and Accurate Distance-Based Reconstruction of Single-Cell Copy Number Phylogenies

Abstract

Somatic copy number alterations (sCNAs) are valuable phylogenetic markers for inferring evolutionary relationships among tumor cell subpopulations. Advances in single-cell DNA sequencing technologies are making it possible to obtain such sCNAs datasets at ever-larger scales. However, existing methods for reconstructing phylogenies from sCNAs are often too slow for large datasets. Moreover, the accuracies of many existing methods are highly sensitive to error and other features of the analyzed datasets. In this work, we propose two new distance-based approaches for reconstructing single-cell tumor phylogenies from sCNA data. The new methods, DICE-bar and DICE-star, are based on novel, easy-to-compute distance measures and drastically outperform the current state-of-the-art in terms of both accuracy and scalability. Using carefully simulated datasets, we find that DICE-bar and DICE-star significantly improve upon the accuracies of existing methods across a wide range of experimental conditions and error rates while simultaneously being orders of magnitude faster. Our experimental analysis also reveals how noise/error in copy number inference, as expected for real datasets, can drastically impact the accuracies of many existing methods. We apply DICE-star, the most accurate method on error-prone datasets, to two real single-cell breast cancer datasets and find that it helps identify previously unreported rare cell populations.