Robust CNV detection using single-cell ATAC-seq

Copy number variation (CNV) is a widely studied type of structural variation seen in the genomes of cancerous and other dysfunctional cells. CNVs can have direct and indirect effects on gene dosage, and are thought to drive cancer progression and other disorders. Advancements in single-cell assays such as sc-ATAC-seq and sc-RNA-seq, along with their ubiquitous use, allows for the identification of CNVs at single cell resolution. While there are a variety of available tools for CNV detection in sc-RNA-seq, development of sc-ATAC-seq based accurate and reliable CNV callers is in the early stages, with only two available algorithms so far. We present RIDDLER, a single-cell ATAC-seq CNV detection algorithm based on outlier aware generalized linear modeling. By utilizing tools from robust statistics, we developed an extensible model that is able to identify single-cell CNVs from sc-ATAC-seq data in an unsupervised fashion, while providing probabilistic justification for results. Our statistical approach also allows us to estimate when loss of signal is likely caused by drop-out or a true genome deletion event, as well as predict reliable CNVs without the need for normative reference cells. We demonstrate the effectiveness of our algorithm on cancer cell line models where it achieves better agreement with bulk WGS derived CNVs than competing methods. We also compare our approach on 10x multimone data, where it shows better agreement and integration with RNA derived CNV estimates.