[3Xin2] Poster session 1

[3Xin2-17] Drug Target Discovery using Machine Learning on Literature-based Knowledge Graph

〇Mariko Nio¹, Shiori Nakazawa², Koki Kibe², Tomoyuki Namai¹, Ryouichi Chatani¹, Manabu Wada² (1.Biometrics Department, Chugai Pharmaceutical Co., Ltd., 2.Biological Technology Department, Chugai Pharmaceutical Co., Ltd.)

Keywords:Knowledge Graph, Knowledge Graph Embeddings, Literature-based Data, Link Prediction

In the selection of drug target genes, research papers involving the relationships between gene knockout/knockdown and disease provide critical information. Biomedical entities such as genes and diseases are interconnected in a network, and it is a general concept to extract this information from literature data. By constructing a massive knowledge graph (KG) from the literature information and predicting new gene-disease relations, we aim to propose drug target genes. In this study, we used the search results by Elsevier Text Mining as data source. Genes, diseases, and their relations were extracted and constructed a KG. The triples derived from papers before 2021 were used as the Train/Valid set, and those from literature after 2022 were used as the Test set. We predicted the new gene-disease relations using a KG completion algorithms. As a result, we were able to predict new relations that were included in the Test set. This suggests that the potential to explore drug target genes from past literature information.