Building a comprehensive knowledge graph for incRNA

Abstract

In the present study, we developed a rich and focused knowledge graph that explores complex relationships between a variety of entities such as genes, diseases, chemicals, and variants related to lncRNAs. This information has been aggregated from thousands of high-quality sources, including PharmGKB, CTD, and DisGeNET, lncRNASNP v3, LncRNADisease v3.0, resulting in a dataset of more than 4.5 million unique relationships. Applying the complete standardization process, Gilda in our case, we curated the identifiers and enriched the data with high-ranking associations. This graph contains both directed and bidirectional relations, enabling the exploration of complex biological interactions or relations. By calibrating uneven degree distributions and combining a combinatorial approach, we identified six critical pairwise relationships, such as Gene-Disease and Chemical-Variant, Chemical-Gene, Chemical Disease, Gene-Variant, Disease-Variant associations, ensuring comprehensive coverage of biomedical knowledge. We developed a shortest path length-based method to predict target genes for long non-coding RNAs (lncRNAs) using a knowledge graph (KG). By extracting all lncRNAs and genes from the KG, we constructed a distance matrix where each entry represented the shortest path length between an lncRNA and a gene, calculated using the Breadth-First Search (BFS) algorithm. Genes with a path score below the mean for a given lncRNA were identified as potential targets. Predictions were validated using benchmark datasets from LncTarD and RNAInter, ensuring reliability and alignment with experimentally validated interactions.