摘要：Next-generation sequencing (NGS) is widely used in genetic testing for the highly sensitive detection of single nucleotide changes and small insertions or deletions. However, detection and phasing of structural variants, especially in repetitive or homologous regions, can be problematic due to uneven read coverage or genome reference bias, resulting in false calls. To circumvent this challenge, a computational approach utilizing customized scaffolds as supplementary reference sequences for read alignment was developed, and its effectiveness demonstrated with two CBS gene variants: NM_000071.2:c.833T>C and NM_000071.2:c.[833T>C; 844_845ins68]. Variant c.833T>C is a known causative mutation for homocystinuria, but is not pathogenic when in cis with the insertion, c.844_845ins68, because of alternative splicing. Using simulated reads, the custom scaffolds method resolved all possible combinations with 100% accuracy and, based on > 60,000 clinical specimens, exceeded the performance of current approaches that only align reads to GRCh37/hg19 for the detection of c.833T>C alone or in cis with c.844_845ins68. Furthermore, analysis of two 1000 Genomes Project trios revealed that the c.[833T>C; 844_845ins68] complex variant had previously been undetected in these datasets, likely due to the alignment method used. This approach can be configured for existing workflows to detect other challenging and potentially underrepresented variants, thereby augmenting accurate variant calling in clinical NGS testing.
