Abstract
A recent paper by Jiang et al. in BMC Bioinformatics presented guidelines on long-read sequencing settings for structural variation (SV) calling, and benchmarked the performance of various SV calling tools, including NanoVar. In their simulation-based benchmarking, NanoVar was shown to perform poorly compared to other tools, mostly due to low SV recall rates. To investigate the causes for NanoVar's poor performance, we regenerated the simulation datasets (3× to 20×) as specified by Jiang et al. and performed benchmarking for NanoVar and Sniffles. Our results did not reflect the findings described by Jiang et al. In our analysis, NanoVar displayed more than three times the F1 scores and recall rates as reported in Jiang et al. across all sequencing coverages, indicating a previous underestimation of its performance. We also observed that NanoVar outperformed Sniffles in calling SVs with genotype concordance by more than 0.13 in F1 scores, which is contrary to the trend reported by Jiang et al. Besides, we identified multiple detrimental errors encountered during the analysis which were not addressed by Jiang et al. We hope that this commentary clarifies NanoVar's validity as a long-read SV caller and provides assurance to its users and the scientific community.