Abstract

The current practice of treating cancer is a one-size-fits-all approach, in which patients with the same type and stage of cancer receive the same treatment. This approach is ineffective 75% of the time. With microRNA (miRNA) having been identified as a key biomarker of cancer, precision therapeutics based on miRNA should provide the highest specificity and sensitivity by virtue of their cancer-specific expression and stability. However, identifying particular miRNAs that play a key role in driving cancer remains a challenge, as the expression of some types of miRNAs is found to be significantly different between normal tissues and tumor tissues. The focus of this research is to create a pan-cancer solution using machine learning to identify key miRNAs as biomarkers of cancer, and predict drug combinations based on miRNA. Data from 23 cancer types from The Cancer Genome was used. Top miRNAs were identified as key biomarkers using ExtraTreesClassifier, and were validated through functional enrichment analysis and survival analysis. Three different models were implemented using Multi-label ML algorithms: K-NearestNeighbors, AdaBoostClassifier, and OneVsRestClassifier, with OneVsRestClassifier yielding the highest accuracy. The final model was tuned using cross validation and a novel Median Scoring Method, based on F1 Score, Jaccard Score, and Accuracy Score. The resulting solution overcomes the challenges of monotherapy and allows oncologists to prescribe anti-cancer drug combinations with high accuracy based on patients' miRNAs, yielding higher survivability.