Deep learning-based splicing prediction of rare variants in Dilated Cardiomyopathy

Continuously decreasing costs for sequencing has led to a rising number of patients receiving routine genetic testing as part of the clinical workup. Consequently, the number of variants to be classified according to the 5-tier classification system of the American College of Medical Genetics (ACMG) is also rising, leading to a substantial fraction of variants of unknown significance (VUS). Advanced computational methods are now available to help in a more accurate predication of a possible variant outcome.

This project has aimed to validate deep-learning-based variant effect predictions of the spliceAI-tool, by analysing DNA sequencing alongside with RNA-seq from the same individuals with dilated cardiomyopathy (DCM).

For this analysis, we used two cohorts with available genome sequencing and RNA-sequencing data, where one cohort of 50 DCM patients had polyA-based RNA-seq from routine biopsies from heart catheterizations as well as RNA-seq from peripheral blood and an independent cohort with 35 DCM patients, where samples have been taken from end-stage lv-explants. SpliceAI-scores for donor/acceptor gain/loss were annotated according to Jaganathan et. al. and junction reads were counted with leafcutter.

In both cohorts, we were able to confirm an increasing validation rate according to a higher spliceAI-score culminating in the most stringent spliceAI-score bin (0.8-1), reaching >75% for single nucleotide variants (SNPs) predicted to result in an acceptor/donor gain. Validation rate in blood was lower, not reaching 70%. From 14205 potential junction sites covered in both tissue and blood, 4926 (35%) could be validated cross tissue. Among those,we find variants previously classified as VUS to result in splicing alterations in important cardiac genes like LMNA or TNNC1. Cascade genetic testing could identify additional family members carrying the same variant, co-segregating with the disease.

In summary, our study supports the added value of incorporating advanced computational tools such as spliceAi in the genetic workup of routine genetic testing in dilated cardiomyopathy. If available, the addition of personalized RNA-seq analysis in addition to DNA-based genetic testing should be considered if potential splice-variants have been identified.