In silico prediction analysis

To predict the impact on protein function of the nonsynonymous variants identified, we applied the following computer algorithms, which are currently recommended by the ACMG/AMP guidelines: PANTHER⁴⁷, MutationTaster⁴⁸, Condel⁴⁹, PROVEAN⁵⁰, PolyPhen2⁵¹, Sort Intolerant from tolerant (SIFT)⁵², Align Grantham Variation/Grantham Difference score (GVGD)⁵³, Combined Annotation Dependent Depletion (CADD)⁵⁴, PhD-SNPg⁵⁵, Functional Analysis through Hidden Markov Models (FATHMM)⁵⁶, SNPs&GO⁵⁷, and MutPred2 (http://mutpred.mutdb.org).

For Align-GVGD, we classified the variants based on the graded classifier, with a cutoff of C35 or higher for deleterious classification. For CADD, we used the PHRED-like score with a cutoff of 20, below which variants were classified as benign and otherwise deleterious. For MutPred2, we considered a score threshold of 0.50 for pathogenicity. For all other algorithms, we considered the classification provided as output.

We used the DynaMut server (http://biosig.unimelb.edu.au/dynamut/) to assess the impact of mutations on protein dynamics and stability⁵⁸. The server requires an input file of protein structure in PDB format or by providing the four-letter accession code for any entry at the Protein Data Bank database (PDB; http://wwpdb.org). The used code for the FURIN gene was 5jxg. The other proteins are not available in PDB.