Comprehensive Summary
This article evaluated multiple computational methods to measure reliability and accuracy in predicting cancer drivers. For known cancer drivers, the authors found that multi-computational tools incorporating both functional genomic and protein structure data alongside evolutionary data outperformed predictions compared to models trained solely on evolutionary data. In a population of non-small cell lung cancer patients divided into two cohorts, the researchers determined the validity of AI predictions of cancer drivers with unknown significance (VUs) by testing their association with the overall survival rate in the population. AI was able to identify certain VUs such as KEAP1 and SMARCA4 as pathogenic drivers, a prediction that was associated with worse survival in the population of cancer patients.
Outcomes and Implications
These findings indicate a novel method for identifying the patient-outcomes of VUs in a more accurate manner than currently available. If employed in a clinical setting, medical intervention can be employed early-on to address both the predicted rate of progression and the predicted symptoms associated with this rate. The computational method takes into account multiple contributing factors towards a particular cancer driver’s likelihood of being a pathogenic driver, providing a more comprehensive understanding of a given patient’s survival outcome. As a result, health-care providers can use such a computational method to best diagnose, treat, and predict the health outcomes of cancer patients, potentially improving quality of life alongside quantity of life.