Researcher Furukawa from the Discovery Technologies Business Unit to present a poster at the 53rd Annual Meeting of the Japanese Society of Toxicology (JSOT)

2026.06.16

Researcher Furukawa from the Discovery Technologies Business Unit to present a poster at the 53rd Annual Meeting of the Japanese Society of Toxicology (JSOT),  to be held at Osaka International Convention Center, Japan.
If you plan to attend the conference, we encourage you to join the session and hear the presentation.

Conference Name: the 53rd Annual Meeting of the Japanese Society of Toxicology (JSOT)

Date & Time: July 2 (Thu), 2026, 13:30-14:20 (JST)

Location: Osaka International Convention Center, Japan

Title: Predicting Drug-Induced Liver Injury Using Machine Learning: A Novel Approach Integrating Multiple In Vitro Assays

Poster Board No: P155

Presentation Summary
Drug-induced liver injury (DILI) is widely recognized as a major cause of drug development discontinuation and post-marketing withdrawal. Therefore, early prediction of its risk is of critical importance. However, its underlying mechanisms are diverse and complex, making it difficult to evaluate using a single assay.
In this study, prior to the application of machine learning, we first evaluated the predictive performance of five parameters derived from four types of assays conducted during the lead optimization stage—cytotoxicity (Cytotox IC50/Cmax), mitochondrial toxicity (Glu/Gal ratio), cholestasis (BSEP IC50/Cmax), phospholipidosis (relative value to the AUC of amiodarone), and reactive metabolite formation (production rate)—together with Cmax values obtained from the literature, using ROC analysis for individual assessment.
Next, a machine learning model was constructed based on six parameters—Cytotox IC50, Glu/Gal ratio, BSEP IC50, relative value to the AUC of amiodarone, reactive metabolite formation rate, and Cmax—together with molecular descriptor information calculated from SMILES, using 70 compounds for training, and the predictive performance of the integrated risk assessment was evaluated.
The results showed that cholestasis exhibited the highest predictive accuracy in the individual analysis. Furthermore, validation of the machine learning model using 16 compounds demonstrated a sensitivity of 81.8% and a specificity of 80.0% for DILI risk. In addition, as this approach reflects the influence of Cmax, it enabled the prediction of blood concentrations associated with DILI risk based on the calculated risk scores even for compounds with unknown Cmax values.
In conclusion, it was demonstrated that the machine learning model constructed based on five parameters and molecular descriptor information enables the prediction of DILI risk.

Axcelead DDP’s Soulution
Drug-induced liver injury (DILI) is a major safety concern in drug development, and early prediction of its risk is expected to mitigate potential adverse outcomes.
We achieve DILI risk prediction by integrating data obtained at the lead optimization stage with molecular descriptors, together with machine learning approaches that incorporate Cmax (maximum plasma concentration), an indicator of compound exposure.