Table 1.

Classification of methods to predict the effect of missense mutations

Method type	Prediction	Limitations
Protein stability	Predicts the difference in unfolding free energy between wild-type and mutant protein	Considers only one possible mechanism that may affect the phenotype
Protein–protein/protein–nucleic acid affinity	Predicts the difference in the binding affinity between binding partners upon mutation	Small training datasets limit the scope of these methods
Protein–ligand affinity	Predicts the difference in ligand-binding affinity upon mutation	Small training datasets limit the scope of these methods
Phenotypic effect	Predicts the likelihood that a mutation is deleterious without considering a specific molecular mechanism	Except for Mendelian disease phenotypes, the phenotype may only be observed in a subset of the population (partial penetrance). Databases use different annotation practices and contain contradictory information for some mutations
Mapping and 3D visualization	Provides a 3D context of the site of mutation and may give atomic-level insight into mechanism of action	Visual approach is not suitable for automated whole-exome predictions
3D mutation hotspots	Clusters mutations by spatial proximity that are not necessarily close in protein sequence	Clustering may not explain the effect of specific mutations in a hotspot

3D three-dimensional