ELE: Estimating tissue-specific long noncoding RNA gene essentiality using graph neural networks

A gene is essential if its loss of function results in lethality, reduced fitness, or disease. Essential genes have attracted increasing attention due to their vital functions in biological systems. Many efforts have been made to find essential protein-coding genes. In complex organisms, the majority of the genome does not encode proteins. Noncoding genes play vital roles in many biological processes. Investigating the essentiality of noncoding genes is necessary. We present ELE (estimating long noncoding RNA essentiality), the first computational model for finding essential long noncoding RNA (lncRNA) genes in a tissue-specific context. ELE integrates lncRNA-protein-protein interaction (LPPI) network structure with genomic and epigenomic features to predict essential lncRNA genes. With ELE, we identified tissue-specific essential lncRNA genes in human and mouse. Tissue-specific essential lncRNA genes are more involved in physiological context-dependent cellular functions. Moreover, we found that essential lncRNA genes tend to interact more with essential proteins and essential microRNAs (miRNAs). ELE serves as an effective framework for deciphering lncRNA gene essentialities, offering a set of tissue-dependent candidates for further validation. We have deposited all datasets and codes for reproducing ELE results in a GitHub repository (swt1024/ELE).