Personalized Federated Learning and AI for Accelerating Drug Discovery

In this presentation, Huawei senior engineer Xu Chi introduces his research on computer‑aided drug design and new drug development, focusing on the challenges of drug screening and key issues in the discovery pipeline.

He discusses how artificial intelligence, including supervised learning, reinforcement learning, generative models, and graph computing, has transformed drug discovery, enabling rapid data integration, target prediction, and biomarker identification.

The session highlights successful AI applications such as protein 3D structure prediction (iPhord), drug repurposing, and the AutoOmics platform for biomarker discovery, as well as large‑scale virtual screening powered by Huawei Cloud's 15,000‑core compute resources.

A major focus is personalized federated learning, a distributed machine‑learning approach that protects proprietary pharmaceutical data while allowing collaborative model training. Xu describes the FedAMP algorithm, which improves over FedAvg by weighting contributions based on model similarity and data quality.

He also outlines the workflow for launching a drug‑focused federated learning service on Huawei ModelArts, including federation creation, participant invitation, agent deployment, and result monitoring, and presents cloud‑based virtual screening services that visualize binding affinities across thousands of compounds.