Network-Based Computational Modeling to Unravel Gene Essentiality

Essential genes are reductively defined as those fundamental for an organism’s reproductive success and growth. Still, the so-called essentiality of a gene is a context-dependent dynamic attribute that can vary in different cells, tissues, or pathological conditions. Identifying essential genes at a genome-wide level is a challenging issue in primary and applied biomedical research, prominently in synthetic biology, drug targeting, and disease gene identification. Wet-lab experimental procedures designed to test whether a gene is essential or not are cost- and time-consuming, especially in the case of complex organisms such as humans. Consequently, computational approaches provide a fundamental alternative, still representing a demanding and challenging task due to the complex nature of the biological problem. Commonly explored methods are devoted to classifying nodes in protein-protein interaction networks, but they are scarcely successful, especially in the case of human genes. Node classification in graph modeling/analysis allows predicting an unknown node property based on defined node attributes. Here, we propose an overview of the different aspects of the biological background, methodologies, and applications related to identifying essential genes, with the aim to provide a small guide through the potentialities and open issues. We further present an experimental approach to examine the entire workflow, from the labeling of the nodes to the attribute choice to the learning modeling. To this extent, we exploit a tissue-specific integrated network enriched with pre-computed biological and embedding-derived topological features to develop a model through a deep learning approach.