Drug induced Liver Injury Prediction with Injective Molecular Transformer

Geonyeong Choi; Hyo Jung Cho; Soon Sun Kim; Ji Eun Han; Jaeyoun Cheong; Charmgil Hong

Abstract

Drug-Induced Liver Injury (DILI), liver damage caused by drugs, represents a significant factor contributing to the failure of clinical trials. Remarkably, the drug development process, which entails an extensive timeline spanning several years and incurring costs of billions of dollars to achieve FDA approval, could greatly benefit from early DILI prediction. Furthermore, through the utilization of DILI prediction, clinicians can obtain valuable insights into the potential risks associated with medication, empowering them to make more informed decisions when prescribing drugs to patients. We employed Graph Neural Networks (GNNs) to predict DILI based on drug structures. GNNs consist of node aggregation, which gathers node representations, and graph pooling, which compiles node representations to portray the graph as a single vector. The graph pooling method built on Set Transformer outperforms existing techniques, but we identified a limitation: Set Transformer, using a random seed vector as the query vector, cannot differentiate between graphs of varied structures. Moreover, it was found to potentially lack expressiveness, being randomly defined without prior knowledge and relying on a limited number of seed vectors. To overcome this issue, we introduced Molecular Transformer which employs the unique molecular representation as the query vector. We found that using drug toxicity information extracted from drug toxicity knowledge-bases as the query vector yielded the best performance.

Keywords: Drug induced Liver Injury; Graph Neural Networks; Classification; Graph Pooling; Molecular Transformer

Links
[Full text PDF][Bibtex]