Machine learning optimized DriverDetect software for high precision prediction of deleterious mutations in human cancers - Scientific Reports
Summary
The F1 score combines the parameters of precision and recall to provide insights into the specificity and sensitivity of the program during the learning process, even in the face of issues like class swapping44. Finally, MCC provides a reliable evaluation of the machine learning performance in the event of an unbalanced dataset of multi-class cases by denoting the number of samples categorized in one label as bigger than another. However, VEST (CRAVAT version 5.2.4, https://www.cravat.us/CRAVAT/) has a lower overall accuracy (87.5%), indicating a slightly poorer general performance, primarily due to fewer true negatives in its prediction. This approach is also demonstrated in ensemble tools where authors have combined multiple simpler and smaller algorithms, each focusing on different aspects of mutation detection (for example, amino acid conservation or chemical bond disruption), to produce a more comprehensive tool21. To this point, DriverDetect attained an accuracy of 93%, an F1 score of 0.9286, and an MCC of 0.8607, which overall provided a better performance than MutPred (version 2.0, http://mutpred.mutdb.org/; best single population-based tool) or even the top combination, [PredictSNP + VEST].