A New Computational Approach to Identify Essential Genes in Bacterial Organisms Using Machine Learning

摘要

Essential genes of an organism are those genes that are required for the growth to a fertile adult and is pivotal for the survival of an organism. In this study, a new computational approach based on machine learning method is designed which can constructively project essential genes by integration of homologous, gene intrinsic, and network topology features. A set of 15 bacterial organisms as reference species have been used which have characterized essential genes. By applying "Extreme Gradient Boosting (XGBoost)" for Bacillus Subtilis 168, the classification model through tenfold cross-validation test gave average AUC value of 0.9649. Further applying this new model to a closely related organism Salmonella enterica serovar Typhimurium LT2 resulted in a very definitive AUC value of 0.8608. To assess the stability and consistency of the proposed classifier, a different set of target organisms comprised of Escherichia coli MG1655 and Streptococcus sanguinis SK36 and another classifier based on PCR method were implemented. The performance of the model based on principal component regression (PCR) method for both set of target organisms resulted in lower AUC values. It shows that the newly designed feature-integrated approach based on XGBoost method results in better predictive accuracy to identify essential genes.