Context: The interpretation of novel missense variants is a challenge with increasing numbers of such variants being identified and a responsibility to report the findings in the context of all available scientific evidence. Various in silico bioinformatic tools have been developed that predict the likely pathogenicity of missense variants; however, their utility within the diagnostic setting requires further investigation.
Aim: The aim of our study was to test the predictive value of two of these tools, sorting intolerant from tolerant (SIFT) and polymorphism phenotyping (PolyPhen), in a set of 141 missense variants (131 pathogenic, 8 benign) identified in the ABCC8, GCK, and KCNJ11 genes.
Methods: Sixty-six of the mutations caused a gain of protein function, while 67 were loss-of-function mutations. The evolutionary conservation at each residue was also investigated using multiple sequence alignments from the UCSC genome browser.
Results: The sensitivity of SIFT and PolyPhen was reasonably high (69% and 68%, respectively), but their specificity was low (13% and 16%). Both programs were significantly better at predicting loss-of-function mutations than gain-of-function mutations (SIFT, p = 0.001; PolyPhen, p < or = 0.0001). The most reliable method for assessing the likely pathogenicity of a missense variant was to investigate the degree of conservation at the affected residue. Eighty-eight percent of the mutations affected highly conserved residues, while all of the benign variants occurred at residues that were polymorphic across multiple species.
Conclusions: Although SIFT and PolyPhen may be useful in prioritizing changes that are likely to cause a loss of protein function, their low specificity means that their predictions should be interpreted with caution and further evidence to support/refute pathogenicity should be sought before reporting novel missense changes.