Developing Predictive Models by Sharing Predictions - An Investigation of a Federated Learning Approach for ADMET Predictions
Clinical Snapshot
PICO Framework
| P — Population | Pharmaceutical compounds requiring ADMET (Absorption, Distribution, Metabolism, Excretion, Toxicity) predictions, specifically rat steady-state volume of distribution data from 11 pharmaceutical companies |
| I — Intervention | Student-teacher model (STM) federated learning framework where organizations share predictions on public datasets to create pseudolabeled training data |
| C — Comparator | Individual teacher models from participating organizations and existing federated learning approaches (MELLODY and Effiris) |
| O — Outcomes | Model performance measured by RMSE on external test sets, workflow simplicity, data privacy preservation, and scalability for cross-company collaboration |
Bottom Line
This proof-of-concept study demonstrates that a student-teacher federated learning model can achieve comparable predictive performance to individual pharmaceutical company models while preserving data privacy. The approach successfully aggregated predictions from 11 companies on ~133,000 compounds for rat steady-state volume of distribution, achieving RMSE of 0.51 versus 0.47-0.61 for individual models. The simplified workflow offers practical advantages over existing federated learning methods by avoiding direct data sharing and iterative collaboration requirements. While promising for pharmaceutical ADMET prediction, the study is limited to a single endpoint and lacks detailed statistical validation. The methodology could significantly benefit Australian drug development by enabling secure collaboration between local and international pharmaceutical partners while maintaining competitive advantages through proprietary data protection.
Key Findings
P Value: Not reported
Effect Size: Student model RMSE ≈ 0.51 compared to teacher models range 0.47-0.61
Primary Outcome: Student model performance on external test set measured by RMSE
Nnt Or Sensitivity: Performance comparable to individual teacher models with simplified workflow
Confidence Interval: Not clearly reported
Clinical Application
High feasibility given simpler workflow compared to traditional federated learning approaches, requiring only prediction sharing rather than iterative collaboration Highly relevant for Australian pharmaceutical research institutions and TGA regulatory science, particularly for companies collaborating internationally on drug development while maintaining proprietary data security Pharmaceutical researchers developing ADMET prediction models, particularly relevant for drug discovery and development teams
Abstract
Machine learning models for ADMET prediction benefit from large, diverse data sets, yet such data are typically siloed across organizations. Federated learning (FL) enables collaborative modeling while preserving data privacy. Here, we investigate a student-teacher model (STM) framework in which organizations train internal models on proprietary data and share predictions on a public data set to generate pseudolabels for a centralized student model. As a proof of concept, 11 pharmaceutical companies contributed predictions for rat steady-state volume of distribution, yielding a pseudolabeled data set of ∼133,000 compounds. The resulting student model achieved performance comparable to individual teacher models on an external test set (RMSE ≈ 0.51 vs 0.47-0.61). Compared with FL approaches such as MELLODY and Effiris, STM offers a simpler workflow that avoids direct data sharing or iterative collaboration, providing a practical and scalable framework for secure cross-company model development.
References
- 1.Guha, R., Wang, W., Price, E., Bhhatarai, B., Hassan, M., DiFranzo, A., Keefer, C., Woody, N., Broccatelli, F., Winiwarter, S., He, L., Green, D. V. S., Edwards, C. D., Kato, H., Kosugi, Y., & Desai, P. (2026). Developing predictive models by sharing predictions - An investigation of a federated learning approach for ADMET predictions. Journal of Medicinal Chemistry. https://doi.org/10.1021/acs.jmedchem.5c03681
Related Research
CPT: pharmacometrics & systems pharmacology
Structured Schemas for Provenance-Rich, LLM-Assisted QSP Model Calibration
2 Aug 2026
Journal of pharmacokinetics and pharmacodynamics
Diffusion models for virtual populations and pharmacometric simulations
1 Aug 2026
Journal of chemical information and modeling
HyperDC: A Non-Uniform Hypergraph Framework for Dual- and Higher-Order Drug Combination Recommendation Across Diverse Complex Diseases
29 July 2026
This content is for educational purposes for healthcare professionals only and does not constitute clinical advice. Clinical decisions should be based on individual patient assessment, current guidelines, and appropriate specialist consultation. Editorial Standards · Privacy Policy · Terms of Service