Performance of meta-predictors for the classification of MED13L missense variations, implication of raw parameters

MED13L syndrome is a rare congenital disorder comprising moderate intellectual disability, hypotonia and facial dysmorphism. Whole exome or genome sequencing in patients with non-specific neurodevelopmental disorders leads to identification of an increasing number of MED13L missense variations of unknown signification. The aim of our study was to identify relevant annotation parameters enhancing discrimination between candidate pathogenic or neutral missense variations, and to assess the performance of seven meta-predictor algorithms: BayesDel, CADD, DANN, FATHMM-XF, M-CAP, MISTIC and REVEL for the classification of MED13L missense variants. Significant differences were identified for five parameters: global conservation through verPhyloP and verPhCons scores; physico-chemical difference between amino acids estimated by Grantham scores; conservation of residues between MED13L and MED13 protein; proximity to phosphorylation sites for pathogenic variations. Among the seven selected in-silico tools, BayesDel, REVEL, and MISTIC provided the most interesting performances to discriminate pathogenic from neutral missense variations.