Cost-effective federated molecular design with FedMol

Tiejun Dong; Linlin You; Calvin Yu-Chian Chen

doi:10.59717/j.xinn-drugdisc.2026.100022

The Innovation Drug Discovery > 2026 Vol. 1 > No. 2 > 100022

Next Article Previous Article

ARTICLE Open Access

Cite PDF

Cost-effective federated molecular design with FedMol

1.
Intelligent Medical Research Center, School of Intelligent Systems Engineering, Sun Yat-sen University, Shenzhen 518107, China
2.
School of AI for Science, Peking University Shenzhen Graduate School, Shenzhen 518055, China
3.
State Key Laboratory of Chemical Oncogenomics, Key Laboratory of Chemical Genomics, Peking University Shenzhen Graduate School, Shenzhen 518055, China
4.
Guangdong Provincial Key Laboratory of Intelligent Transportation System, School of Intelligent Systems Engineering, Sun Yat-sen University, Shenzhen 510275, China

More Information

Corresponding authors: youllin@mail.sysu.edu.cn (L.Y.); cy@pku.edu.cn (C.Y.C.)
Received Date: January 21, 2026

Accepted Date: June 02, 2026

Published Online: June 11, 2026

The Innovation Drug Discovery 1(2), https://doi.org/10.59717/j.xinn-drugdisc.2026.100022 | Cite this article

GRAPHICAL ABSTRACT

GRAPHICAL ABSTRACT

DownLoad: Full size image

PUBLIC SUMMARY

PUBLIC SUMMARY
1. A federated framework for collaborative training of molecular generative models.
  
  Combining low rank adaptation (LoRA) with knowledge distillation (KD).
  
  FedMol achieves superior performance under high data heterogeneity, maintaining better distribution matching.

ABSTRACT

ABSTRACT

Generating molecules with properties of interest is a fundamental challenge in computational molecular design. However, a significant portion of valuable chemical data resides within proprietary corporate data silos, limiting the potential of data-driven approaches for molecular design. Federated learning provides a framework for collaborative model training without sharing raw data, but existing methods face challenges in communication efficiency and robustness to heterogeneous data distributions. Here we introduce FedMol, a federated learning framework that combines pre-trained molecular language models with parameter-efficient fine-tuning (using low-rank adaptation) and knowledge distillation to enable privacy-preserving molecular generation. We evaluate FedMol on the GuacaMol distribution-learning benchmarks under varying degrees of data heterogeneity simulated via Dirichlet partitioning. Our results show that FedMol achieves competitive performance with state-of-the-art centralized models by updating only 2.4% of parameters through low-rank adaptation. Under high heterogeneity, knowledge distillation outperforms standard federated averaging in preserving chemical diversity. This work demonstrates a practical approach to collaborative molecular design that respects data privacy, with potential applications in drug discovery where proprietary data and multi-institutional collaboration are both critical.
- molecular generation /
- federated learning /
- knowledge distillation

FullText(HTML)

REFERENCES

[1]	Kang, N.Y., Ha, H.H., Yun, S.W. et al. (2011). Diversity-driven chemical probe development for biomolecules: beyond hypothesis-driven approach. Chem. Soc. Rev. 40:3613−3626. DOI:10.1039/C0CS00172D View in Article CrossRef Google Scholar
[2]	Honarparvar, B., Govender, T., Maguire, G.E. et al. (2014). Integrated approach to structure-based enzymatic drug design: molecular modeling, spectroscopy, and experimental bioactivity. Chem. Rev. 114:493−537. DOI:10.1021/cr300314q View in Article CrossRef Google Scholar
[3]	Bagal, V., Aggarwal, R., Vinod, P. et al. (2021). MolGPT: molecular generation using a transformer-decoder model. J. Chem. Inf. Model. 62:2064−2076. DOI:10.1021/acs.jcim.1c00600 View in Article CrossRef Google Scholar
[4]	Gómez-Bombarelli, R., Wei, J.N., Duvenaud, D. et al. (2018). Automatic chemical design using a data-driven continuous representation of molecules. ACS Cent. Sci. 4:268−276. DOI:10.1021/acscentsci.7b00572 View in Article CrossRef Google Scholar
[5]	Li, C. (2025). Breaking data silos in drug discovery with federated learning. Nat. Chem. Eng. 2:288−289. DOI:10.1038/s44286-025-00213-x View in Article CrossRef Google Scholar
[6]	Durant, G., Boyles, F., Birchall, K. et al. (2024). The future of machine learning for small-molecule drug discovery will be driven by data. Nat. Comput. Sci. 4:735−743. DOI:10.1038/s43588-024-00699-0 View in Article CrossRef Google Scholar
[7]	Hanser, T., Ahlberg, E., Amberg, A. et al. (2025). Data-driven federated learning in drug discovery with knowledge distillation. Nat. Mach. Intell. 1−14. DOI:10.1038/s42256-025-00991-2 View in Article CrossRef Google Scholar
[8]	Chen, S., Xue, D., Chuai, G. et al. (2020). FL-QSAR: a federated learning-based QSAR prototype for collaborative drug discovery. Bioinformatics 36:5492−5498. DOI:10.1093/bioinformatics/btaa1006 View in Article CrossRef Google Scholar
[9]	Li, L., Fan, Y., Tse, M. et al. (2020). A review of applications in federated learning. Comput. Ind. Eng. 149:106854. DOI:10.1016/j.cie.2020.106854 View in Article CrossRef Google Scholar
[10]	Huang, T., Xu, H., Wang, H. et al. (2023). Artificial intelligence for medicine: Progress, challenges, and perspectives. Innov. Med. 1:100030. DOI:10.59717/j.xinn-med.2023.100030 View in Article CrossRef Google Scholar
[11]	Zhu, W., Luo, J., and White, A.D. (2022). Federated learning of molecular properties with graph neural networks in a heterogeneous setting. Patterns 3. DOI:10.1016/j.patter.2022.100521. DOI:10.1016/j.patter.2022.100521 View in Article CrossRef Google Scholar
[12]	Cozac, R., Hasic, H., Choong, J.J. et al. (2025). kMoL: an open-source machine and federated learning library for drug discovery. J. Cheminform. 17:22. DOI:10.1186/s13321-025-00967-9 View in Article CrossRef Google Scholar
[13]	Bassani, D., Brigo, A., and Andrews-Morger, A. (2023). Federated learning in computational toxicology: an industrial perspective on the effiris hackathon. Chem. Res. Toxicol. 36:1503−1517. DOI:10.1021/acs.chemrestox.3c00137 View in Article CrossRef Google Scholar
[14]	Luo, B., Li, X., Wang, S. et al. (2021). Cost-effective federated learning design. In IEEE INFOCOM 2021. pp. 1–10. DOI: 10.1109/INFOCOM42981.2021.9488679. View in Article Google Scholar
[15]	Zhu, Z., Hong, J., and Zhou, J. (2021). Data-free knowledge distillation for heterogeneous federated learning. In ICML. pp. 12878–12889. DOI: 10.48550/arXiv.2105.10056. View in Article Google Scholar
[16]	Lu, Z., Pan, H., Dai, Y. et al. (2024). Federated learning with non-IID data: A survey. IEEE Internet Things J. 11:19188−19209. DOI:10.1109/JIOT.2024.3376548 View in Article CrossRef Google Scholar
[17]	Hu, E.J., Shen, Y., Wallis, P. et al. (2022). LoRA: Low-rank adaptation of large language models. In ICLR. pp. 3. DOI: 10.48550/arXiv.2106.09685. View in Article Google Scholar
[18]	Gou, J., Yu, B., Maybank, S.J. et al. (2021). Knowledge distillation: A survey. Int. J. Comput. Vis. 129:1789−1819. DOI:10.1007/s11263-021-01453-z View in Article CrossRef Google Scholar
[19]	Brown, N., Fiscato, M., Segler, M.H. et al. (2019). GuacaMol: benchmarking models for de novo molecular design. J. Chem. Inf. Model. 59:1096−1108. DOI:10.1021/acs.jcim.8b00839 View in Article CrossRef Google Scholar
[20]	Yang, A., Yang, B., Hui, B. et al. (2024). Qwen2 technical report. ArXiv. DOI: 10.48550/arXiv.2407.10671. View in Article Google Scholar
[21]	Ainslie, J., Lee-Thorp, J., De Jong, M. et al. (2023). GQA: Training generalized multi-query transformer models from multi-head checkpoints. In Proc. EMNLP 2023. pp. 4895–4901. DOI: 10.18653/v1/2023.emnlp-main.298. View in Article Google Scholar
[22]	Su, J., Ahmed, M., Lu, Y. et al. (2024). Roformer: Enhanced transformer with rotary position embedding. Neurocomputing 568:127063. DOI:10.1016/j.neucom.2023.127063 View in Article CrossRef Google Scholar
[23]	Li, T., Sahu, A.K., Zaheer, M. et al. (2020). Federated optimization in heterogeneous networks. In Proc. Mach. Learn. Syst. pp. 429–450. DOI: 10.48550/arXiv.1812.06127. View in Article Google Scholar
[24]	Kim, S., Chen, J., Cheng, T. et al. (2023). PubChem 2023 update. Nucleic Acids Res. 51:D1373−D1380. DOI:10.1093/nar/gkac956 View in Article CrossRef Google Scholar
[25]	Gaulton, A., Bellis, L.J., Bento, A.P. et al. (2012). ChEMBL: a large-scale bioactivity database for drug discovery. Nucleic Acids Res. 40:D1100−D1107. DOI:10.1093/nar/gkr777 View in Article CrossRef Google Scholar
[26]	Shokri, R., Stronati, M., Song, C. et al. (2017). Membership inference attacks against machine learning models. In Proc. IEEE Symp. Secur. Priv. pp. 3–18. DOI: 10.1109/SP.2017.41. View in Article Google Scholar
[27]	He, Y., Li, B., Liu, L. et al. (2025). Towards Label-Only membership inference attack against pre-trained large language models. In Proc. 34th USENIX Secur. Symp. pp. 1609–1628. DOI: 10.48550/arXiv.2502.18943. View in Article Google Scholar
[28]	Preuer, K., Renz, P., Unterthiner, T. et al. (2018). Fréchet chemnet distance: a metric for generative models for molecules in drug discovery. J. Chem. Inf. Model. 58:1736−1741. DOI:10.1021/acs.jcim.8b00234 View in Article CrossRef Google Scholar
[29]	Cavanagh, J.M., Sun, K., Gritsevskiy, A. et al. (2024). Smileyllama: Modifying large language models for directed chemical space exploration. arXiv. DOI: 10.48550/arXiv.2409.02231. View in Article Google Scholar
[30]	Izdebski, A., Olszewski, J., Gawade, P. et al. (2025). Synergistic benefits of joint molecule generation and property prediction. arXiv. DOI: 10.48550/arXiv.2504.16559. View in Article Google Scholar
[31]	Chen, X., Wang, Y., He, J. et al. (2025). Graph generative pre-trained transformer. arXiv. DOI: 10.48550/arXiv.2501.01073. View in Article Google Scholar
[32]	Kwon, Y., Yoo, J., Choi, Y. et al. (2019). Efficient learning of non-autoregressive graph variational autoencoders for molecular graph generation. J. Cheminform. 11:70. DOI:10.1186/s13321-019-0396-x View in Article CrossRef Google Scholar
[33]	Iwata, H., Nakai, T., Koyama, T. et al. (2023). VGAE-MCTS: A new molecular generative model combining the variational graph auto-encoder and monte carlo tree search. J. Chem. Inf. Model. 63:7392−7400. DOI:10.1021/acs.jcim.3c01220 View in Article CrossRef Google Scholar
[34]	Wu, J.N., Wang, T., Chen, Y. et al. (2024). t-SMILES: a fragment-based molecular representation framework for de novo ligand design. Nat. Commun. 15:4993. DOI:10.1038/s41467-024-49388-6 View in Article CrossRef Google Scholar
[35]	Mahmood, O., Mansimov, E., Bonneau, R. et al. (2021). Masked graph modeling for molecule generation. Nat. Commun. 12:3156. DOI:10.1038/s41467-021-23415-2 View in Article CrossRef Google Scholar
[36]	Geng, Z., Xie, S., Xia, Y. et al. (2023). De novo molecular generation via connection-aware motif mining. arXiv. DOI: https://doi.org/10.48550/arXiv.2302.01129. View in Article Google Scholar
[37]	Huang, Y., Zeng, Y., Dwivedi, V.P. et al. (2026). Higher-order grammar representations for molecular generation and learning. In ICLR. View in Article Google Scholar
[38]	Ueda, R., Nakai, T., Yoshida, K. et al. (2025). Mitigation of membership inference attack by knowledge distillation on federated learning. IEICE Trans. Fundam. Electron. Commun. Comput. Sci. 108:267−279. DOI:10.1587/transfun.2024CIP0004 View in Article CrossRef Google Scholar
[39]	Karimireddy, S.P., Kale, S., Mohri, M. et al. (2020). Scaffold: Stochastic controlled averaging for federated learning. In ICML. pp. 5132–5143. DOI: 10.48550/arXiv.1910.06378. View in Article Google Scholar
[40]	You, J., Ying, R., Ren, X. et al. (2018). Graphrnn: Generating realistic graphs with deep auto-regressive models. In ICML. pp. 5708–5717. DOI: 10.48550/arXiv.1802.08773. View in Article Google Scholar
[41]	Schütt, K., Kindermans, P.J., Sauceda Felix, H.E. et al. (2017). Schnet: A continuous-filter convolutional neural network for modeling quantum interactions. vol. 30. DOI: 10.48550/arXiv.1706.08566. View in Article Google Scholar
[42]	Satorras, V.G., Hoogeboom, E., and Welling, M. (2021). E(n) equivariant graph neural networks. In ICML. pp. 9323–9332. DOI: 10.48550/arXiv.2102.09844. View in Article Google Scholar
[43]	Stärk, H., Beaini, D., Corso, G. et al. (2022). 3d infomax improves gnns for molecular property prediction. In ICML. pp. 20479–20502. DOI: 10.48550/arXiv.2110.04126. View in Article Google Scholar
[44]	Carlini, N., Chien, S., Nasr, M. et al. (2022). Membership inference attacks from first principles. In Proc. IEEE Symp. Secur. Priv. pp. 1897–1914. DOI: 10.1109/SP46214.2022.9833649. View in Article Google Scholar
[45]	Shi, S., Zhang, J., Jiang, S. et al. (2025). DP-GENG: Differentially private dataset distillation guided by dp-generated data. DOI: 10.1609/aaai.v40i11.37854. View in Article Google Scholar
[46]	Abadi, M., Chu, A., Goodfellow, I. et al. (2016). Deep learning with differential privacy. In ACM CCS. pp. 308–318. DOI: 10.1145/2976749.2978318. View in Article Google Scholar
[47]	Bonawitz, K., Ivanov, V., Kreuter, B. et al. (2017). Practical secure aggregation for privacy-preserving machine learning. In ACM CCS. pp. 1175–1191. DOI: 10.1145/3133956.3133982. View in Article Google Scholar

Cited by

ABOUT THIS ARTICLE

Cite this article:

Dong T., You L. and Chen C. (2026). Cost-effective federated molecular design with FedMol. The Innovation Drug Discovery 1:100022. https://doi.org/10.59717/j.xinn-drugdisc.2026.100022

Dong T., You L. and Chen C. (2026). Cost-effective federated molecular design with FedMol. The Innovation Drug Discovery 1:100022. https://doi.org/10.59717/j.xinn-drugdisc.2026.100022

Welcome!

To request copyright permission to republish or share portions of our works, please visit Copyright Clearance Center's (CCC) Marketplace website at marketplace.copyright.com.

PDF

Figures(4) Tables(5)

Article Metrics

Article views(436) PDF downloads(142)

Cost-effective federated molecular design with FedMol

GRAPHICAL ABSTRACT

PUBLIC SUMMARY

ABSTRACT

REFERENCES

ABOUT THIS ARTICLE

Cite this article:

Welcome!

Share

Article Metrics

Relative Articles

Cited by

Catalog

{{newsColumn.name}}

Official Account

{{newsColumn.name}}

Stay connected

{{newsColumn.name}}

Cost-effective federated molecular design with FedMol

GRAPHICAL ABSTRACT

PUBLIC SUMMARY

ABSTRACT

REFERENCES

ABOUT THIS ARTICLE

Cite this article:

Welcome!

Share

Article Metrics

Relative Articles

Cited by

Catalog

Export File

Citation

Format

Content

{{newsColumn.name}}

Official Account

{{newsColumn.name}}

Stay connected

{{newsColumn.name}}