MI2026 – Yuwei Zhou

Title:
Enhancing polyreactivity prediction of preclinical antibodies through fine-tuned protein language models

Abstract:
Therapeutic monoclonal antibodies (mAbs) are widely used for treating diverse diseases, yet their clinical success is often hindered by polyreactivity caused by non-specific binding to off-target biomolecules. Early identification of polyreactivity risk is therefore essential to improve developability and reduce failure rates.We developed PolyXpert, a polyreactivity prediction model based on fine-tuned protein language models. Among six evaluated models, PolyXpert achieved the best performance, reaching an AUC of 0.9672 and ~90% accuracy on an independent test set.Our results show that fine-tuning PLMs significantly improves robustness compared to embedding-based approaches. PolyXpert provides an efficient in silico tool for early-stage screening and selection of therapeutic antibody candidates.Code is available at: https://github.com/zzyywww/PolyXpert

Biography:
Yuwei Zhou received her Ph.D. in Biomedical Engineering from the University of Electronic Science and Technology of China. Her research focuses on bioinformatics, particularly AI-driven therapeutic antibody development. She has developed antibody-related databases, computational screening algorithms, and analytical platforms for preclinical drug discovery. She has participated in multiple national and provincial research projects and collaborated with industry on AI-based antibody design targeting plasma kallikrein. She has published nearly 20 papers in journals such as eLife, Journal of Pharmaceutical Analysis, and Methods, with strong expertise in protein sequence analysis and machine learning.