Abstract
Next-generation DNA sequencing (NGS) in short-read mode has recently been used for genetic testing in various clinical settings. NGS data accuracy is crucial in clinical settings, and several reports regarding quality control of NGS data, primarily focusing on establishing NGS sequence read accuracy, have been published thus far. Variant calling is another critical source of NGS errors that remains unexplored at the single-nucleotide level despite its established significance. In this study, we used a machine-learning-based method to establish an exome-wide benchmark of difficult-to-sequence regions at the nucleotide-residue resolution using 10 genome sequence features based on real-world NGS data accumulated in The Genome Aggregation Database (gnomAD) of the human reference genome sequence (GRCh38/hg38). The newly acquired metric, designated the 'UNMET score,' along with additional lines of structural information from the human genome, allowed us to assess the sequencing challenges within the exonic region of interest using conventional short-read NGS. Thus, the UNMET score could provide a basis for addressing potential sequential errors in protein-coding exons of the human reference genome sequence GRCh38/hg38 in clinical sequencing.