Federated learning for molecular discovery.

Area of Science:

• Computational chemistry and cheminformatics
• Machine learning and artificial intelligence
• Drug discovery and development

Background:

• Federated Learning (FL) facilitates collaborative machine learning across decentralized datasets.
• FL mitigates privacy risks inherent in data sharing, crucial for sensitive domains.
• Molecular informatics, particularly drug discovery, is increasingly leveraging FL.

Purpose of the Study:

• To review current projects and applications of Federated Learning in molecular discovery.
• To highlight the benefits of FL in enhancing drug discovery models.
• To identify and discuss the challenges hindering the full exploitation of FL in this field.

Main Methods:

• Systematic review of publications on Federated Learning in molecular discovery.
• Analysis of reported benefits, including model performance and applicability domain.
• Identification of common challenges and limitations across studies.

Main Results:

• Federated Learning demonstrably improves the performance of predictive models in molecular discovery.
• Knowledge aggregation through FL expands the applicability domain of these models.
• Studies confirm the value of FL in collaborative drug discovery efforts.

Conclusions:

• Federated Learning offers significant advantages for molecular discovery by enhancing model performance and enabling secure collaboration.
• Further research is needed to address existing challenges for widespread adoption.
• FL is poised to revolutionize drug discovery through advanced, privacy-preserving AI.