Exploring the impact of sequence context on errors in SNP genotype calling with whole genome sequencing data using AI-based autoencoder approach.

A critical step in the analysis of whole genome sequencing data is variant calling. Despite its importance, variant calling is prone to errors. Our study investigated the association between incorrect single nucleotide polymorphism (SNP) calls and variant quality metrics and nucleotide context. In our study, incorrect SNPs were defined in 20 Holstein-Friesian cows by comparing their SNPs genotypes identified by whole genome sequencing with the IlluminaNovaSeq6000 and the EuroGMD50K genotyping microarray. The dataset was divided into the correct SNP set (666 333 SNPs) and the incorrect SNP set (4 557 SNPs). The training dataset consisted of only the correct SNPs, while the test dataset contained a balanced mix of all the incorrectly and correctly called SNPs. An autoencoder was constructed to identify systematically incorrect SNPs that were marked as outliers by a one-class support vector machine and isolation forest algorithms. The results showed that 59.53% (±0.39%) of the incorrect SNPs had systematic patterns, with the remainder being random errors. The frequent occurrence of the CGC 3-mer was due to mislabelling a call for C. Incorrect T instead of A call was associated with the presence of T in the neighbouring downstream position. These errors may arise due to the fluorescence patterns of nucleotide labelling.