Disease Prediction using Optimal Feature Selection from Epigenetic Data

Epigenetic changes are identified to correlate with gene expression changes among numerous diseases including cancer. Although integrated data analysis tools are available, a quantitative model that precisely predicts the diseases from epigenetic information is still in its infancy. In this paper, a data mining approach is presented for predicting a disease using the most relevant genes (treated as features) associated with it. A few existing feature selection methods have been applied independently and their results combined to form a fused feature set. For selecting the optimal set of genes from them, a genetic algorithm based evolutionary approach is proposed. They have been applied to a classifier for recognizing the disease from a set of reference samples. A key observation from our analysis is that the optimization using genetic algorithm gives better prediction accuracy than the existing methods. Hence the proposed optimal fusion based feature selection can be effectively applied for better disease diagnosis and prognosis.