Table 2. Amino acid substitution (AAS) prediction methods used in this study.

Program	Input	Algorithm	Output	URL	Reference
SIFT	PS and AAS, protein sequence alignment and AAS, dbSNP id, or protein id	Uses sequence homology, scores assessment is based on position-specific scoring matrices with Dirichlet priors	Score ranges from 0 to 1, where < = 0.05 is damaging and >0.05 is tolerated	http://sift.jcvi.org/www/SIFT_enst_submit.html	Ng and Henikoff, 2001 [63]
PolyPhen-2	PS and AAS, dbSNP id, HGVbASE id, or protein id	Uses sequence conservation and structure to model location of amino acid substitution, Swiss-Prot and TrEMBL annotation	Score ranges from 0 to 1, where < = 0.05 is benign, and >0.05 is damaging	http://genetics.bwh.harvard.edu/pph2/	Ramensky et al. 2002 [50]
PANTHER-PSEP	PS and AAS	Uses sequence homology; scores are based on PANTHER Hidden Markov Model families	Probably damaging: time > 450my possibly damaging: 450my > time > 200my probably benign: time < 200my)	http://www.pantherdb.org/tools/csnpScoreForm.jsp	Tang and Thomas, 2016 [64]
MutPred	Protein id, PS, or multiple sequence alignment	Prediction is based on one of two neural networks which uses internal databases, secondary structure prediction, and sequence conservation	Score ranges from 0 to 1, where 0 is polymorphism and high scores are predicted to be deleterious/disease-associated	http://mutpred.mutdb.org/	Li et al. 2009 [65]
MutatioTaster	DNA sequence	Predictions are calculated by a naive Bayes classifier, which predicts the disease potential	Prediction is based one of four possible types: a) disease causing: probably deleterious b) disease causing automatic: known to be deleterious c) polymorphism: probably harmless d) polymorphism automatic: known to be harmless	http://www.mutationtaster.org/	Schwarz et al. 2014 [53]
Provean	PS and AAS	Uses an alignment-based score approach to generate predictions not only for single amino acid substitutions, but also for multiple amino acid substitutions, and in-frame insertions and deletions	the default score threshold is currently set at -2.5, in which >-2.5 is neutral, and <-2.5 is deleterious	http://provean.jcvi.org/index.php	Choi and Chan, 2015 [54]
PMUT	PS and AAS, dbSNP, Uniprot or PDB ID of protein	Based on the application of neural networks which uses internal databases, secondary structure prediction, and sequence conservation	Score ranges from 0 to 1, where <0.50 is neutral and >0.50 is disease associated	http://mmb.pcb.ub.es/pmut2017/analyses/new/	Ferrer-Costa et al. 2002 [55]
FATHMM	protein identifier and the amino acid substitution, dbSNP id	Uses sequence homology	The score threshold is set at -2.5, in which >-2.5 is neutral, and <-2.5 is deleterious	http://fathmm.biocompute.org.uk/index.html	Shihab et al. 2013 [56]
nsSNPAnalyzer	Protein sequence in FASTA format and a substitution file denoting the SNP identities to be analyzed	Uses information contained in the multiple sequence alignment and information contained in the three-dimensional protein structure to make predictions.	Normalized probability of the substitution calculated by the SIFT program	http://snpanalyzer.uthsc.edu/	Bao et al. 2005 [57]
Align GV-GD	Protein sequence in FASTA format and a substitution file denoting the SNP identities to be analyzed	Uses biophysical features of amino acids and protein multiple sequence alignments	A value of C > 0 was considered deleterious; otherwise a variant was neutral	http://agvgd.hci.utah.edu/	Tavtigian et al. 2006 [58]
REVEL	Precomputed REVEL scores are provided for all possible human missense variants	Prediction is based on a combination of scores from 13 individual tools	Score ranges from 0 to 1, where <0.50 is neutral and >0.50 is pathogenic	https://sites.google.com/site/revelgenomics/	Ioannidis et al. 2016 [59]

AAS; amino acid sequences, PS; protein sequence, PDB, protein data bank