Knowledge-Driven Pesticide Repurposing Via Link Prediction with Pesticide Graph Embedding

Abstract

Pesticides are essential for controlling agricultural pests and diseases and increasing crop yields. However, the development of new pesticides requires significant resources and time, leading to a shortage in pesticide supply. Traditional pesticide design methods are heavily reliant on field trials and bioassays for experimental screening, often lacking systematic guidance. To address this limitation, we first propose a novel pesticide repurposing method based on knowledge graph embedding (KGE) and link prediction inspired by drug repurposing. A comprehensive pesticide knowledge graph is constructed and used for training the KGE model to capture the semantic and structural information of the graph by embedding matrix. By applying pesticide-disease link prediction techniques, A potential new relationships can be identified  between pesticides and diseases. This approach can effectively generalize to unseen pesticide-disease relationships, providing a scientific foundation and motivation for biochemical experiments in pesticide repurposing.  Codes and data are available at: http://pesticide-repurposing.samlab.cn.

Keywords: knowledge graph, Knowledge Graph Embedding, Pesticide Repurposing, Link prediction