PTENpred: A Designer Protein Impact Predictor for PTEN-related Disorders

Abstract

Connecting a genotype with a phenotype can provide immediate advantages in the context of modern medicine. Especially useful would be an algorithm for predicting the impact of nonsynonymous single-nucleotide polymorphisms in the gene for PTEN, a protein that is implicated in most human cancers and connected to germline disorders that include autism. We have developed a protein impact predictor, PTENpred, that integrates data from multiple analyses using a support vector machine algorithm. PTENpred can predict phenotypes related to a human PTEN mutation with high accuracy. The output of PTENpred is designed for use by biologists, clinicians, and laymen, and features an interactive display of the threedimensional structure of PTEN. Using knowledge about the structure of proteins, in general, and the PTEN protein, in particular, enables the prediction of consequences from damage to the human PTEN gene. This algorithm, which can be accessed online, could facilitate the implementation of effective therapeutic regimens for cancer and other diseases.

Toward Systems Pathology for PTEN Diagnostics

“…Techniques such as support vector machine learning will analyze data from global datasets to achieve robust predictive algorithms to assist clinicians in advising patients on optimal treatment decisions. An example of these types of analyses is "PTENpred" that uses support vector machine learning to bring together data to predict phenotype based on a PTEN mutation (Johnston and Raines 2016). As additional, well-annotated large datasets become available, the accuracy of these types of approaches will increase.…”