Classification of miRNA expression data using random forests for cancer diagnosis

Cancer is a major leading cause of death and responsible for around 13% of all deaths world-wide. Cancer incidence rate is growing at an alarming rate in Malaysia and the world as we know it. It is estimated that statistically one out of every four Malaysians will develop cancer by the age of 75. Conventional methods of diagnosing cancer rely solely on skilled physicians, with the help of medical imaging, to detect certain symptoms which usually appear in the late stage of cancer. Furthermore, biopsy examinations are highly invasive since tissue samples are required to be extracted from patients. There exist minimally invasive cancer biomarkers in forms of proteins from serum. Nevertheless, existing protein-based diagnosis techniques require labor-intensive analysis compounded by low diagnosis sensitivity. There have indeed been a number of studies to identify novel miRNA-based cancer biomarkers. However, the existing diagnosis techniques using miRNA suffer from low diagnosis accuracy, sensitivity, and specificity. The low diagnosis accuracy and sensitivity of the existing techniques stems from the fact that there is extremely low miRNA count in body fluids. There is also an inevitable problem of cross contamination between cells and exosomes in sample preparation steps. This paper proposes to circumvent these problems in data analysis stage with a machine learning technique called Random Forest. The proposed system achieved 93.48 % accuracy for gastric cancer and 100 % accuracy for ovarian cancer. The results are promising and encouraging. Despite much noise contaminated the sample preparation process and low miRNA count in body fluids, the proposed system able to identify miRNA markers responsible for classification of cancer