A method of identifying false positives in the variety-specific variant calling of rice

Abstract This study investigated the effect of variety (or subspecies) specificity on genetic variant calling using next-generation sequencing data from rice. Specifically, we used two major rice genome varieties, Indica and Japonica, to construct different variant calling models with varying compositions of samples from these two varieties. Our investigation revealed that divergence of samples from the reference sequence corresponded to increased variant prediction. Specifically, including samples that differed in variety from the reference sequence significantly increased the number of variants predicted. We used machine learning techniques to understand this phenomenon and evaluated the performance of different variant calling models based on the predicted variants. Our results indicated that a significant proportion of the additional predicted variants represented potential false positives, which was particularly accentuated when phylogenetically distinct accessions from the reference were included in the samples. To improve the accuracy of the predicted variants, we proposed a method to identify false positives and allow their exclusion if necessary. This proposed approach involved calling true variants from purebred (or typical) samples. We validated the effectiveness of this method across different variant calling models and demonstrated a significant reduction in false-positive predicted variants. As a practical application, we implemented the method on dbSNP of rice, a database of known rice variants, and demonstrated a means to identify false positives within dbSNP. Our study provides general recommendations for best practices in variety-specific variants calling for rice.