Cancer relapse prediction from microrna expression data using machine learning

Cancer is a major deadliest disease globally that involve uncontrolled cell growth and invasion-metastasis events. It accounts for around 13% of all deaths worldwide. Statistical reports have pointed out that the cancer occurrence rate is increasing at an alarming rate in the world. Furthermore, cancer relapse rate is also rising mostly due to late cancer diagnosis. Some cancers can recur at the site of origin or the distant site after years of anti-cancer treatment. Therefore, cancer relapse prediction process is of paramount important so that early specific treatments can be sought. Nevertheless, conventional methods for diagnosing cancer relapse rely on invasive and labor intensive biopsy examinations. Circulating miRNAs have gained great interest in medical field because of their higher sensitivity, specificity and potential for minimally invasive sampling procedures. Furthermore, miRNA expression profiling from body fluid samples using high-throughput approaches is a promising technology that could predict cancer relapse. This paper describes a machine learning based approach called one-dependent estimator to predict cancer relapse from miRNA expression data. The proposed framework will predict whether a particular cancer will relapse within cancer recurrence time frame, which is usually 5 years. To select relevant cancer recurrence associated miRNAs, we employ an entropy-based miRNA marker selection approach. This proposed system has achieved an average accuracy of 92.82% in predicting cancer relapse over three datasets, namely glioblastoma, ovarian cancer, and hepatocellular carcinoma (HCC). The experimental results exhibit the efficacy of the proposed framework