Evaluating Structural Variation Detection Tools for Long-Read Sequencing Datasets in Saccharomyces cerevisiae

Structural variation (SV) represents a major form of genetic variations that contribute to polymorphic variations, human diseases, and phenotypes in many organisms. Long-read sequencing has been successfully used to identify novel and complex SVs. However, comparison of SV detection tools for long-read sequencing datasets has not been reported. Therefore, we developed an analysis workflow that combined two alignment tools (NGMLR and minimap2) and five callers (Sniffles, Picky, smartie-sv, PBHoney, and NanoSV) to evaluate the SV detection in six datasets of Saccharomyces cerevisiae. The accuracy of SV regions was validated by re-aligning raw reads in diverse alignment tools, SV callers, experimental conditions, and sequencing platforms. The results showed that SV detection between NGMLR and minimap2 was not significant when using the same caller. The PBHoney was with the highest average accuracy (89.04%) and Picky has the lowest average accuracy (35.85%). The accuracy of NanoSV, Sniffles, and smartie-sv was 68.67%, 60.47%, and 57.67%, respectively. In addition, smartie-sv and NanoSV detected the most and least number of SVs, and SV detection from the PacBio sequencing platform was significantly more than that from ONT (p = 0.000173).