SOSCNV: a stochastic outlier selection algorithm-based method for copy number variation detection using hybrid strategy

Background Copy number variation (CNV) refers to the duplication or deletion of DNA sequences in the genome. It plays an important role in organism development and disease occurrence. Methods Read-depth (RD)–based strategies are among the most widely used approaches for CNV detection; however, RD signals typically allow only approximate breakpoint localization. In contrast, split-read (SR) methods can provide much more precise breakpoint resolution. As a result, hybrid RD+SR strategies have become increasingly popular in recent CNV detection frameworks. Nevertheless, these hybrid methods still exhibit limitations, particularly in detecting CNVs of diverse lengths, where performance may degrade for very short or very long variants. A new CNV detection method called SOSCNV (Detection of Copy Number Variations based on Stochastic Outlier Selection) is proposed in this article. In order to enhance the recognition and processing ability of those difficult-to-detect areas, and also to make the detection of breakpoints more accurate, SOSCNV incorporates a two-stage segmentation technique. SOS algorithm is used to model the RD signal and Guanine and Cytosine (GC) content, and calculate the correlation between datasets to identify anomalous datasets. Compared with traditional RD+SR hybrid detection strategies, SOSCNV is better suited for identifying CNVs across a wide range of variant lengths and demonstrates stronger robustness under varying sequencing depths and tumor purities. Results After testing on 900 simulated datasets and four real datasets, and comparing with four existing CNV detection tools, experiments show that SOSCNV can significantly improve detection precision, recall and F1-score.