[3Xin2] Poster session 1

[3Xin2-71] Oncology Drug Regimen Prediction using Machine Learning for Knowledge Graph

〇Koyo Tate¹, Yuichi Yano¹, Kenichiro Akagi¹, Yayoi Natsume^2,3, Masataka Kuroda^2,4, Mari Itoh², Chihiro Higuchi², Kenya Ueno⁵, Yi-An Chen², Mariko Nio⁶ (1.NEC Corporation, 2.AI Center for Health & Biomedical Research, National Institutes of Biomedical Innovation, Health and Nutrition, 3.Institute of Advanced Medical Sciences, Tokushima University, 4.Discovery Technology Laboratories, Mitsubishi Tanabe Pharma Corporation, 5.Toray Industries, Inc., 6.Chugai Pharmaceutical Co., Ltd.)

Keywords:knowledge graph, knowledge graph embedding, link prediction, oncology, regimen

As cancer is a complex disease, its treatment is also complex. Patients generally receive multiple line of therapy composed of single or multiple drugs. The strength of drug relationships or synergies have been predicted for doublet in the previous studies. However, it is challenging to predict the relationship or synergies of triplet and more. In this study, we assessed the applicability of machine learning techniques for knowledge graph(KG) for new treatment proposal using drug combination data used in previous clinical trials. Data was extracted from ClinicalTrials.gov(CT.gov) via AACT database and TargetMine(TM). A KG that composed of drug, regimen, disease and gene nodes and their relation were created, and new relation of drug-regimen were predicted using KBLRN, one of the KG link prediction algorithms. The prediction results with solely CT.gov included anticancer drugs in higher ranks. On the other hand, with the addition of TM, drugs other than anticancer drugs were also predicted. This suggests that the potential to propose new regimens not predictable by humans by increasing the diversity of input data.