Predicting pathogenicity of missense mutations

CREATED: 200808140308 Speaker: Shamil Sunyaev ** Genetic diagnostics

• genetic data may inform clinical decisions
• diagnostic by sequencing
• large number of Variants of Unknown Significance (VUS)
• bioinformatics method contribute with diagnostic of VUS if prediction accuracy would increase ** Rare variants in complex phenotypes
• important to predict which variants are functional
• rare variants can be used to predict new genes from all exon sequence data ** Evolutionary genomics
• want to predict mutations of strong effect
• abundance of weakly deleterious alleles in humans
• weakly deleterious variants occur in highly conserved regions ** Predicting effect of missense mutations
• most pathogenic mutations affect stability
• hypermutability is a useful feature for predicting deleterious mutations
• forward and backward selection, five-fold cross-validation, accuracy measured as AUC
• naive bayes with discretization outperformed decision tree and SVM ** Training and testing
• benign mutations: substitutions between close species, human polymorphism
• damaging mutations