The experimentally obtained functional impact assessments of GT>GC 5' splice site variants differ markedly from those predicted

GT>GC 5 splice site (or +2T>C) variants have been frequently reported to cause human genetic disease. However, although we have demonstrated that GT>GC variants in human disease genes may not invariably be pathogenic, none of the currently available splicing prediction tools appear to be capable of reliably distinguishing those GT>GC variants that generate wild-type transcripts from those that do not. Recently, SpliceAI, a novel deep residual neural network tool, has been developed for splicing prediction. Methodologically distinct from previous approaches that either rely on human-engineered features and/or which focus on short nucleotide windows adjoining exon-intron boundaries, SpliceAI assesses splicing determinants by evaluating 10,000 nucleotides of flanking contextual sequence to predict the functional role in splicing of each position in the pre-mRNA transcript. Herein, we evaluated the performance of SpliceAI in the context of three datasets of GT>GC variants, all of which had been characterized functionally in terms of their impact on mRNA splicing. The first two datasets refer to our recently described "in vivo" dataset of 45 disease-causing GT>GC variants and the "in vitro" dataset of 103 GT>GC substitutions. The third dataset comprised 12 BRCA1 GT>GC variants that were recently analyzed by saturation genome editing. We processed all GT>GC variants using the default settings of SpliceAI. Comparison of the SpliceAI-predicted and experimentally obtained functional impact assessments of the analyzed GT>GC variants revealed that although SpliceAI performed rather better than other prediction tools, it was still far from perfect. A key issue is that the impact of GT>GC (as well as GT>GA or +2T>A) variants that generated wild-type transcripts represents a quantitative change that can vary from barely detectable to almost full expression of wild-type transcripts, with wild-type transcripts often co-existing with aberrantly spliced transcripts. Our findings highlight the challenges that we still face in attempting to accurately identify splice-altering variants.