Xuanbai Ren

I am a Ph.D. candidate at the school of computer science, Hunan University, under the supervision of Prof.Xiangxiang Zeng and Prof.Yiping Liu. Previously, I obtained my M.E. degree at the school of computer science, Hunan University in 2022, working with Prof.Lijun Cai and Prof.Xiangzheng Fu. I earned my B.E. degree from Hengyang Normal University in 2019.

Research Interest: My research focuses on AI-driven molecular and peptide discovery, with an emphasis on language models and multi-objective optimization.

Machine Learning: 1) Language Models, studying how model parameters shape generated molecular properties; 2) Multi-objective Optimization, balancing competing objectives in molecular and peptide design.

Publication & Activities: My research has been published in interdisciplinary journals and computer science conferences, including Nature Communications, Advanced Science, Bioinformatics, ACL, AAAI, NeurIPS, and CVPR. A full publication list can be found on Google Scholar.

Our deepAMP study identified the antimicrobial peptide T2-9, whose in vitro activity against E. coli matched that of the clinical antibiotic levofloxacin (MIC: 2.93 μg/mL for both). The designed peptides are covered by a granted Chinese invention patent arising from collaborative research.

Our HydrogelFinder study identified 9 hydrogel-forming peptides among 17 synthesized candidates (52.9%), completing the design-to-experiment workflow in 19 days. A HydrogelFinder-designed gel, PD-DP-008 (JIJ02), has entered Phase I clinical evaluation (CTR20250456).

I also serve as a reviewer for:

Conferences: ICLR (2024-2027), NeurIPS (2025-2026), AAAI (2026-2027)
Journals: Bioinformatics, Briefings in Bioinformatics, Scientific Reports.

More Things: Honored to be recognized as a Top Reviewer at NeurIPS 2026.

news

Oct 16, 2026	🎉 Honored to be recognized as a Top Reviewer at NeurIPS 2026!
Apr 7, 2026	🎉 Our work CAML (A Conflict-Aware Molecular Language Model Merging Framework for Multi-Constraint Molecular Generation) was accepted by ACL 2026 (Main Conference, Long Paper)! It links language-model parameter composition to generated molecular properties and jointly controls four predicted properties without further model training. ** The published paper is available.
Apr 1, 2026	Our multidisciplinary manuscript, “AI Identifies Compounds that Protect Retinal Ganglion Cells and Optic Nerve in Neurodegeneration by Alleviating ER Stress,” studies an AI-driven screening pipeline using ImageMol to identify neuroprotective candidates targeting endoplasmic reticulum stress, with potential relevance to glaucoma. Nature Communications · Under review: major revision.
Mar 1, 2026	🚀 Historical code release: our earlier work, “Multi-objective generative algorithm enables broad-spectrum antibiotic derivatization,” was open-sourced. This announcement is retained as a historical record; see the current publication and manuscript list for up-to-date titles and submission statuses.
Feb 21, 2026	🎉 Our work STEPH (Sparse Task Vector Mixup with Hypernetworks for Efficient Knowledge Transfer in Whole-Slide Image Prognosis) was accepted by CVPR 2026. It studies efficient knowledge transfer for whole-slide image prognosis. ** The published paper and code are available.
Dec 7, 2025	🎉 Our paper “Self-supervised Blending Structural Context of Visual Molecules for Robust Drug Interaction Prediction” was accepted by NeurIPS 2025! The work uses molecular images for robust drug–drug interaction prediction. Published paper.
Feb 25, 2025	🎉 Our paper “Multi-Objective Molecular Design Through Learning Latent Pareto Set” was accepted by AAAI 2025! We propose a novel framework to approximate the entire Pareto set, empowering researchers to fine-tune molecular properties with ease. 🧪✨ Code
Aug 30, 2024	🎉 Our paper “DeepAMP A Foundation Model Identifies Broad-Spectrum Antimicrobial Peptides against Drug-Resistant Bacterial Infection” is published in Nature Communications! In this study, we developed deepAMP, a peptide language-based deep generative framework designed to expedite the discovery of potent, broad-spectrum antimicrobial peptides (AMPs). By employing a novel sequence degradation strategy and multi-stage fine-tuning, we identified 29 high-potential candidates. These peptides proved effective in treating P. aeruginosa skin wound infections in mouse models with excellent biocompatibility. Notably, the identified peptide T2-9 showed *in vitro activity against E. coli* matching that of the clinical antibiotic levofloxacin, with the same MIC of 2.93 μg/mL**.
May 5, 2024	🎉 Our work HydrogelFinder (A Foundation Model for Efficient Self-Assembling Peptide Discovery Guided by Non-Peptidal Small Molecules) is accepted to Advanced Science. By leveraging non-peptidal small molecules to guide chemical space exploration, we’ve achieved a rapid 19-day workflow from generation to wet-lab validation. 🧪 Our model successfully identified 9 novel self-assembling peptides (1-10 amino acids), including the shortest lipid-peptide documented to date. PD-DP-008 (JIJ02), has entered Phase I clinical evaluation (CTR20250456).

selected publications

CAML: A Conflict-Aware Molecular Language Model Merging Framework for Multi-Constraint Molecular Generation

Xuanbai Ren, Luoda Tan, Pei Liu, Tengfei Ma, Xiangzheng Fu, and 3 more authors

In ACL, 2026

Published · First author · Main Conference, Long Paper · CCF A · Natural Language Processing Conference. Links language-model parameter composition to generated molecular properties and jointly controls four predicted properties without further model training.

HTML Code
A Foundation Model Identifies Broad-Spectrum Antimicrobial Peptides against Drug-Resistant Bacterial Infection

Tingting Li†, Xuanbai Ren†, Xiaoli Luo, Zhuole Wang, Zhenlu Li, and 8 more authors

Nature Communications, 2024

Published · Co-first author · Highly cited paper · 161 citations (Google Scholar, October 2026) · IF: 18.1 (2025) · SCI Zone 1 (TOP), JCR Q1. Lead peptide T2-9 matched levofloxacin against E. coli in vitro (MIC: 2.93 μg/mL for both). Designed peptides are covered by a granted Chinese invention patent arising from collaborative research.

† Equal contribution

HTML Code
HydrogelFinder: A Foundation Model for Efficient Self-Assembling Peptide Discovery Guided by Non-Peptidal Small Molecules

Xuanbai Ren, Jiaying Wei, Xiaoli Luo, Yuansheng Liu, Kenli Li, and 10 more authors

Advanced Science, 2024

Published · First author · SCI Zone 1 (TOP), JCR Q1 · IF: 14.1 (2024) · 51 citations (Google Scholar, October 2026). Nine of 17 synthesized candidates formed hydrogels (52.9%) within a 19-day design-to-experiment workflow. A HydrogelFinder-designed gel, PD-DP-008 (JIJ02), entered Phase I clinical evaluation (CTR20250456).

HTML Code
iEnhancer-XG: Interpretable Sequence-Based Enhancers and Their Strength Predictor

Lijun Cai, Xuanbai Ren, Xiangzheng Fu, Li Peng, Mingyu Gao, and 1 more author

Bioinformatics, 2021

Published · First author (supervisor listed first) · Highly cited paper · 98 citations (Google Scholar, October 2026) · CCF A (2026) · IF: 5.5 (2025).

HTML
Sparse Task Vector Mixup with Hypernetworks for Efficient Knowledge Transfer in Whole-Slide Image Prognosis

Pei Liu, Xiangxiang Zeng, Tengfei Ma, Yucheng Xing, Xuanbai Ren, and 1 more author

In CVPR, 2026

Published · Fifth author · CCF A · Computer Vision Conference.

HTML Code
Self-supervised Blending Structural Context of Visual Molecules for Robust Drug Interaction Prediction

Tengfei Ma, Kun Chen, Yongsheng Zang, Yujie Chen, Xuanbai Ren, and 3 more authors

In NeurIPS, 2025

Published · Fifth author · CCF A · Machine Learning Conference.

HTML
Multi-Objective Molecular Design Through Learning Latent Pareto Set

Yiping Liu, Jiahao Yang, Xuanbai Ren, Xinyi Zhang, Yuansheng Liu, and 3 more authors

In AAAI, 2025

Published · Third author · CCF A · Artificial Intelligence Conference.

HTML
AI Identifies Compounds that Protect Retinal Ganglion Cells and Optic Nerve in Neurodegeneration by Alleviating ER Stress

Rong Rong†, Yi Zhou†, Xuanbai Ren†, Jinglin Zhang, Yixian Liu, and 4 more authors

Nature Communications, 2026

Co-first author · Under review: major revision (October 2026). An ImageMol-based screening pipeline for neuroprotective candidates targeting endoplasmic reticulum stress, with potential relevance to glaucoma. This manuscript has not been accepted.

† Equal contribution

Code