Structural variants (SVs) over 50 base pairs play a significant role in phenotypic diversity and are associated with various diseases, but their analysis is complex and resource-intensive. Numerous computational tools have been developed for SV prioritization, yet their effectiveness in biomedicine remains unclear. Here we benchmarked eight widely used SV prioritization tools, categorized into knowledge-driven (AnnotSV, ClassifyCNV) and data-driven (CADD-SV, dbCNV, StrVCTVRE, SVScore, TADA, XCNV) groups in accordance with the ACMG guidelines. We assessed their accuracy, robustness, and usability across diverse genomic contexts, biological mechanisms and computational efficiency using seven carefully curated independent datasets. Our results revealed that both groups of methods exhibit comparable effectiveness in predicting SV pathogenicity, although performance varies among tools, emphasizing the importance of selecting the appropriate tool based on specific research purposes. Furthermore, we pinpointed the potential improvement of expanding these tools for future applications. Our benchmarking framework provides a crucial evaluation method for SV analysis tools, offering practical guidance for biomedical research and facilitating the advancement of better genomic research tools.
© 2024 Liu et al.