CanPredict

CanPredict predicts whether specific protein sequence changes (missense variants) are likely to be cancer-associated by integrating multiple computational scores and a classifier to distinguish driver mutations from polymorphisms.

Key Features:

• Sorting Intolerant From Tolerant (SIFT): Assesses whether an amino acid substitution affects protein function using sequence conservation.
• Pfam-based LogR.E-value metric: Uses Pfam domain information and a LogR.E-value metric to evaluate the functional impact of mutations within protein domains.
• Gene Ontology Similarity Score (GOSS): Measures similarity between a gene harboring a variant and known cancer-causing genes based on Gene Ontology annotations.
• Random forest classifier: Integrates scores from SIFT, Pfam LogR.E-value, and GOSS to predict whether a mutation is likely cancer-associated.
• Low false-positive rate: Demonstrates a low false-positive rate when distinguishing true cancer-associated mutations from polymorphisms and other missense changes.
• Prediction of activating and inactivating mutations: Predicts both activating and inactivating oncogenic missense mutations as deleterious.
• Distinguishes drivers from passengers: Identifies features that differentiate oncogenic (driver) mutations from passenger variants and common SNPs.
• Utility in large-scale analyses: Applied to large-scale genomic datasets to prioritize recurrently observed mutations that are more likely cancer-associated than rare variants.
• Identification of novel germline changes: Has identified previously unknown germline variants with potential oncogenic significance, including P1104A in TYK2.

Scientific Applications:

• Mutation prioritization for experimental validation: Ranks missense variants to guide selection of candidates for functional studies.
• Driver versus passenger classification: Supports discrimination of causal cancer-associated mutations from passenger variants and polymorphisms.
• Oncogenomics and translational research: Aids identification of candidate cancer-associated mutations to inform studies of tumorigenesis and potential therapeutic targets.

Methodology:

Computes SIFT, a Pfam-based LogR.E-value metric, and a Gene Ontology Similarity Score (GOSS) for each missense variant and combines these scores using a random forest classifier to predict cancer-association.