LMADCNV: A CNV Detection Method Based on Local Features and MAD for NGS Data.

Abstract

Copy number variations (CNVs) are a type of structural variation in the genome that impact gene dosage, with significant implications for both normal phenotypic variability and susceptibility to disease. The existing copy number variation detection methods have unstable sensitivity in data with different coverage depths, and cannot identify shorter copy number variation fragments. In this context, we introduce a new method called LMADCNV, specifically designed for detecting CNVs in single-sample data from next-generation sequencing (NGS). LMADCNV employs local features constructed via a cluster partitioning strategy, in conjunction with an anomaly scoring mechanism predicated on median absolute deviation, to facilitate the detection of CNVs. This innovative approach prudently leverages the positional correlation inherent in read depth (RD) data to achieve increased sensitivity without a significant loss in precision. Empirical validation through simulation and real-sample experiments confirms the superiority of LMADCNV over other seven CNV detection methods. LMADCNV not only offers a novel perspective for extracting local features but also shows promise as a robust and effective tool for CNV detection.