Leveraging Random Forest and Graph-based Centralities to Predict Yeast Essential Genes

This paper presents a novel computational approach to identify essential genes (EGs) based on topological features. EGs are genes that are indispensable for the survival of an organism. Traditionally, EGs have been identified through manual gene knock-out experiments in wet labs. However, this process is time-consuming and labor-intensive. Our proposed approach automates EG identification by applying graph theory concepts to protein-protein interaction (PPI) networks. We compute 15 topological features for each gene in the network, such as degree centrality and betweenness centrality. We then use a Random Forest classifier with backward feature elimination to identify the most influential features for EG prediction. Our experimental results show that our proposed approach achieves a very high area under the curve (AUC) when classifying EGs and non-EGs using extracted features from the DIP and BioGrid Saccharomyces cerevisiae PPI networks.