Mutations in phosphatase and tensin homolog (PTEN) have been associated with many cancers. While many mutations have been identified, with most proposed to be pathogenic, their underlying molecular mechanisms remain unknown. In bridging this gap, I have used a structural-based analysis to guide the phenotypic characterization of novel PTEN mutations encountered in the clinic.
We used the mCSM suite of computational tools to quantitatively assess the impact of 7,244 experimentally characterised PTEN missense mutations on protein stability, dynamics and binding affinities to ligands and interacting proteins. These structural and biophysical changes correlated strongly with the experimental biological fitness scores. They were therefore used to build a predictive classifier to distinguish disease-associated mutations from benign ones.
During model building, the original dataset comprising 7,244 missense mutations was divided into a train set (80%) and non-redundant test set (20%). The algorithm that had a consistent performance among train and test sets was chosen. Properties which highly stratified (p-value <0.05) between malignant and benign mutations were kept as core features during feature selection, and combinations of other features were run to choose the best performing classifier. This tool was used to analyse and structurally characterise PTEN mutations found in the COSMIC database to help identify pathogenic and potential passenger mutations.
Due to the importance of PTEN in cancer, and the current lack of mechanistic understandings, the impact of this work is twofold: an understanding of mutational effects at the protein level, which might explain the resultant phenotype, and the leverage of this initial understanding to build a predictive classifier for clinical use. Subsequently, this work can be used to guide drug development efforts and screening strategies.