Accurate and interpretable ADMET prediction: Integrating structural, geometric, and global molecular context representations

Clinical Snapshot

Key Findings

• P Value: Not reported in abstract

• Effect Size: Not reported in abstract — specific performance metrics (AUC-ROC, RMSE, accuracy, F1) are absent; magnitude of improvement over baselines is unquantifiable from available information

• Primary Outcome: LGSM consistently outperformed standard computational baselines across four public ADMET datasets under scaffold-split evaluation conditions

• Nnt Or Sensitivity: Not applicable (computational benchmarking study) — sensitivity, specificity, and false discovery rates for individual ADMET endpoints are not reported in the abstract; these are critical metrics for assessing practical utility in drug screening

• Confidence Interval: Not reported in abstract

Clinical Application

Feasibility for routine use is uncertain. Multi-view deep learning frameworks incorporating 3D conformational descriptors and transformer-based language model encodings require substantial computational infrastructure. The need for 3D conformer generation adds a preprocessing step absent from simpler 2D-based tools. Integration into existing drug discovery workflows (e.g., Schrödinger, OpenEye, or open-source RDKit pipelines) is not described. Availability as open-source software or web server is not mentioned in the abstract. This study has no direct immediate relevance to Australian clinical practice, PBS listings, or TGA regulatory decisions. Indirect relevance exists at the level of pharmaceutical R&D: Australian academic drug discovery programs (e.g., through MRFF-funded medicinal chemistry initiatives, Monash Institute of Pharmaceutical Sciences, or the ARC Centre of Excellence for Innovations in Peptide and Protein Science) could potentially utilise such tools to improve early-stage compound triage. The TGA's regulatory framework for new chemical entities would ultimately require experimental ADMET data regardless of in silico predictions. RACGP guidelines are not applicable. The tool may be of interest to Australian biotechnology companies engaged in small-molecule drug development. Not directly applicable to patient populations. The tool is intended for use by medicinal chemists and computational chemists during early-phase drug discovery to prioritise compounds with favourable ADMET profiles before synthesis and biological testing. Potential indirect benefit to patients through reduced late-stage drug development failures and improved safety of compounds advancing to clinical trials.

Abstract

Predicting absorption, distribution, metabolism, excretion, and toxicity (ADMET) profiles early in drug discovery is essential to avoid late-stage failures. However, most current computational models depend on single-view molecular representations. This restricts them from capturing the complex interplay between 2D topology, 3D conformations, and specific functional groups that collectively drive structure-pharmacokinetic relationships. Here, we present LGSM, a comprehensive deep learning framework for integrative molecular property prediction. By fusing structural (2D topology), geometric (3D conformational descriptors), and sequence-derived semantic representations, LGSM provides accurate ADMET predictions. Specifically, a language model-based encoding of SMILES sequences is utilized to capture the global molecular context, seamlessly complementing local functional group characteristics. By fusing these different levels of information, our architecture accounts for both local chemical environments and broader spatial relationships. We evaluated this approach across four public ADMET datasets. To mimic real-world drug design scenarios where new chemotypes are constantly explored, we applied a strict scaffold split. Under these conditions, LGSM consistently outperformed standard computational baselines. Beyond predictive accuracy, we prioritized chemical interpretability. By combining the model's internal analysis with molecular docking, we show that LGSM successfully flags key substructures responsible for metabolic liabilities or transporter recognition. This transparency makes the framework a practical tool to guide rational lead optimization in medicinal chemistry.

References

1. 1.Qiu, K., Li, K., Qiao, J., Wang, D., Jin, J., Yan, Z., Wong, K.-C., & Wei, L. (2026). Accurate and interpretable ADMET prediction: Integrating structural, geometric, and global molecular context representations. European Journal of Medicinal Chemistry, 118855. https://doi.org/10.1016/j.ejmech.2026.118855