Apheris logo
Menu

Federated Learning for ADMET Prediction: Expanding Model Applicability

This blog explores state-of-the-art science in ADMET prediction, showing how federated learning enables pharma companies to collaboratively train models on diverse data, achieving higher accuracy and broader applicability without compromising data privacy.

Accurate prediction of absorption, distribution, metabolism, excretion, and toxicity (ADMET) properties remains a fundamental challenge in drug discovery. Despite the progress of graph-based deep learning and foundation models, even the most advanced approaches continue to be constrained by the data on which they are trained. Experimental assays are heterogeneous and often low-throughput, while available datasets capture only limited sections of chemical and assay space. As a result, model performance typically degrades when predictions are made for novel scaffolds or compounds outside the distribution of training data. Recent benchmarking initiatives such as the Polaris ADMET Challenge have made this issue explicit. Multi-task architectures trained on broader and better-curated data consistently outperformed single-task or non-ADMET pre-trained models, achieving up to 40–60% reductions in prediction error across endpoints including human and mouse liver microsomal clearance, solubility (KSOL), and permeability (MDR1-MDCKII). These results highlight that data diversity and representativeness, rather than model architecture alone, are the dominant factors driving predictive accuracy and generalization.

Federated Learning as a Technique for Increasing Data Diversity

Because each organization’s assays describe only a small fraction of the relevant chemical space, isolated modeling efforts remain inherently limited. Federated learning provides a method to overcome this limitation by enabling model training across distributed proprietary datasets without centralizing sensitive data. Cross-pharma research has already provided a consistent picture of the advantages of this approach:

  • Federation alters the geometry of chemical space a model can learn from, improving coverage and reducing discontinuities in the learned representation

  • Federated models systematically outperform local baselines, and performance improvements scale with the number and diversity of participants (Heyndrickx et al., JCIM 2023; Oldenhof et al., AAAI 2023).

  • Applicability domains expand, with models demonstrating increased robustness when predicting across unseen scaffolds and assay modalities (Heyndrickx et al., JCIM 2023; Hanser et al., Nat. Mach. Intell. 2025).

  • Benefits persist across heterogeneous data, as all contributors receive superior models even when assay protocols, compound libraries, or endpoint coverage differ substantially (Heyndrickx et al., 2023; Zhu et al., Nat. Commun. 2022; Cozac et al., J. Cheminf. 2025).

  • Multi-task settings yield the largest gains, particularly for pharmacokinetic and safety endpoints where overlapping signals amplify one another (Heyndrickx et al., 2023; Wenzel et al., JCIM 2019).

Together, these findings suggest that federation systematically extends the model’s effective domain, an effect that cannot be achieved by expanding isolated internal datasets.

Best Practices, Better Models: How Apheris Brings Rigour to the Federated ADMET Network

At Apheris, we share the Polaris team’s conviction that rigorous, transparent benchmarks are the foundation of trustworthy machine learning in drug discovery. That belief guides us from the very first moment data arrives through to model evaluation. That’s why every ADMET model we build follows recommended practices from the paper “Practically Significant Method Comparison Protocols” (Ash, J. R. et al., 2025) to deliver results our partners can truly rely on. For our pre-trained models, we carefully validate datasets (performing sanity and assay consistency checks) with normalisation. Data is then sliced by scaffold, assay, and activity cliffs, ensuring we grasp modelability before training begins. With this solid foundation, we move to modelling, where our ADMET models are trained and evaluated using scaffold-based cross-validation runs across multiple seeds and folds, evaluating a full distribution of results rather than a single score. Finally, the appropriate statistical tests are applied to those distributions to separate real gains from random noise. We benchmark against various null models and noise ceilings, so our partners can clearly see true performance gains. This end-to-end discipline pays off: our pre-trained model competes with the top performers across various Polaris ADMET tasks, highlighting that careful data curation and best-practice modelling translate into real gains. When we fine-tune models to partner datasets using the same rigorous approach, we create tailored models that maximise value and impact for every customer, working with them to assess how performance improvement translates to improved molecule prioritization.

Collaborative Progress Towards Generalizable ADMET Models

Approximately 40–45% of clinical attrition continues to be attributed to ADMET liabilities. As model performance increasingly becomes limited by data rather than algorithms, the ability to learn across distributed proprietary datasets, without compromising data confidentiality or intellectual property, will be central to advancing predictive pharmacology. The Apheris Federated ADMET Network provides a framework through which pharmaceutical organizations can jointly train and evaluate ADMET models, expanding chemical coverage while maintaining complete governance and ownership of their data. Through systematic application of federated learning and rigorous methodological standards, the field moves closer to developing models with truly generalizable predictive power across the chemical and biological diversity encountered in modern drug discovery.

Authors: Lewis Mervin, Calum Hand

Resources:

  • Ash, J. R. et al. (2025). Practically Significant Method Comparison Protocols for Machine Learning in Small-Molecule Drug Discovery. ChemRxiv. 2024 DOI: 10.26434/chemrxiv-2024-6dbwv-v2 Link: https://pubs.acs.org/doi/full/10.1021/acs.jcim.5c01609

  • Cozac, R. et al. (2025). kMoL: an open-source machine and federated learning library for drug discovery. J. Cheminformatics 17, 22. DOI: 10.1186/s13321-025-00967-9 Link: https://jcheminf.biomedcentral.com/articles/10.1186/s13321-025-00967-9

  • Hanser, T. et al. (2025). Data-driven federated learning in drug discovery with knowledge distillation. Nature Machine Intelligence 7, 423–436. DOI: 10.1038/s42256-025-00991-2 Link: https://www.nature.com/articles/s42256-025-00991-2

  • Heyndrickx, W. et al. (2023). MELLODDY: cross-pharma federated learning at unprecedented scale unlocks benefits in QSAR without compromising proprietary information. J. Chem. Inf. Model. 63(7), 2331–2344. DOI: 10.1021/acs.jcim.3c00799 Link: https://pubs.acs.org/doi/10.1021/acs.jcim.3c00799

  • Oldenhof, M. et al. (2022). Industry-Scale Orchestrated Federated Learning for Drug Discovery. arXiv preprint. DOI: 10.48550/arXiv.2210.08871 Link: https://arxiv.org/abs/2210.08871

  • Sun, D., Gao, W., Hu, H. & Zhou, S. (2022). Why 90% of clinical drug development fails and how to improve it? Acta Pharmaceutica Sinica B 12(7), 3049–3062. DOI: 10.1016/j.apsb.2022.02.002 Link: https://doi.org/10.1016/j.apsb.2022.02.002

  • Wenzel, J. et al. (2019). Predictive Multitask Deep Neural Network Models for ADME-Tox Properties from Large Data Sets. J. Chem. Inf. Model. 59(3), 1253–1268. DOI: 10.1021/acs.jcim.8b00785 Link: https://pubmed.ncbi.nlm.nih.gov/30615828/ PubMed

  • Zhu, W. et al. (2022). Federated learning of molecular properties with graph neural networks in a heterogeneous setting (FedChem). Nat. Commun. 13, 3099. DOI: 10.1038/s41467-022-35139-9 Link: https://pubmed.ncbi.nlm.nih.gov/35755872/

Federated learning & analytics
Machine learning & AI
Share blog post to

Insights delivered to your inbox monthly

Related Posts

Improving ADMET model performance by collaborating on proprietary data
Improve the accuracy and applicability of ADMET models through collaborative training on proprietary life sciences data - while preserving IP protection
Privacy-preserving Data Science on QSAR models
This case study shows the effectiveness of Apheris’ platform and services for data science on data that is not directly accessible and distributed. Federated learning decentralizes learning algorithms to access multiple data sets, maintaining privacy. Without pooling sensitive data, new insights can be uncovered securely.

In this white paper, you will learn:

checkmark How federated learning, privacy testing and other tools that enable data science on not directly accessible and distributed data
checkmark Sowcase of federated learning, privacy testing and other tools that enable data science practices on not directly accessible and distributed data, at the example of QSAR models.
Insights into the Unknown: Federated Data Diversity Analysis on Molecular Data
AI methods often rely on limited public datasets, restricting industrial impact. Federated learning enables secure collaboration across private pharma data, but data access remains a major challenge.