Identification of disease-critical genes causing preeclampsia: Meta-heuristic approaches

Abstract

A major cause of maternal and neonatal mortality throughout the world is based on disease like preeclampsia and related hypertensive disorders. Through studies it is identified that the disease has genetic factors and it is hereditary in nature. Searching of the genes responsible for the disease is carried out through different bioinformatics and biotechnological experiments. Here, in this paper, search for the disease critical genes is conducted depending on microarray gene expression data which is obtained as a matrix with rows representing genes and columns (samples) as expression levels of the genes at tissues of different patients. From microarray datasets we have taken 25,000 genes with 75 normal samples and 75 diseased samples. A small subset of genes has to be selected as disease critical. Selecting this small subset of critical genes from 25,000 gene set involves a large search space. So meta-heuristic algorithms like Variable Neighborhood Search (VNS) and Differential Evolution (DE) has been applied and their performance has been compared. Fitness of a set of critical genes (solution) is determined by k Nearest Neighbor (kNN) method. This fitness calculation is performed by classifying each sample with respect to its k (here 3) neighbors. Fitness (number of samples properly classified) is high in both implementations where 80–90 out of 150 samples were classified properly. It is observed that DE outperforms VNS in terms of both average and best fitness values when executed for the same runtime.