Bio AI Agent: A multi-agent artificial intelligence system for autonomous CAR-T cell therapy development with integrated target discovery, toxicity prediction, and rational molecular design
- Bio LIMS INC
- May 29
- 2 min read
Yi Ni*, Liwei Zhu, Shuai Li
Bio LIMS INC, Boston, Massachusetts, United States
*Corresponding author: yi.ni@biolims.net
arXiv preprint
November 11, 2025
q-bio.QM · cs.AI
Abstract
Chimeric antigen receptor T-cell (CAR-T) therapy represents a paradigm shift in cancer treatment, yet development timelines of 8–12 years and clinical attrition rates exceeding 40–60% highlight critical inefficiencies in target selection, safety assessment, and molecular optimization. We present Bio AI Agent, a multi-agent artificial intelligence system powered by large language models that enables autonomous CAR-T development through collaborative specialized agents. The system comprises six autonomous agents: Target Selection Agent for multi-parametric antigen prioritization across >10,000 cancer-associated targets, Toxicity Prediction Agent for comprehensive safety profiling integrating tissue expression atlases and pharmacovigilance databases, Molecular Design Agent for rational CAR engineering, Patent Intelligence Agent for freedom-to-operate analysis, Clinical Translation Agent for regulatory compliance, and Decision Orchestration Agent for multi-agent coordination. Retrospective validation demonstrated autonomous identification of high-risk targets including FcRH5 (hepatotoxicity) and CD229 (off-tumor toxicity), patent infringement risks for CD38+SLAMF7 combinations, and generation of comprehensive development roadmaps. By enabling parallel processing, specialized reasoning, and autonomous decision-making superior to monolithic AI systems, Bio AI Agent addresses critical gaps in precision oncology development and has potential to accelerate translation of next-generation immunotherapies from discovery to clinic.
Key contributions
A six-agent collaborative architecture with specialized roles spanning the entire CAR-T development pipeline from target identification to clinical translation planning.
Integration of heterogeneous data sources including knowledge graphs of >10,000 cancer-associated antigens, human tissue expression atlases (GTEx, Human Protein Atlas), pharmacovigilance databases (FDA FAERS), patent databases, and 50 million PubMed abstracts.
Autonomous toxicity prediction framework validated through retrospective analysis of problematic clinical candidates (FcRH5, CD229) with mechanistic risk profiling.
Intelligent patent landscape analysis enabling freedom-to-operate assessment and competitive positioning strategies.
Generation of comprehensive development roadmaps with short-term (1–3 months), mid-term (3–6 months), and long-term (6–12 months) action plans integrating technical, regulatory, and commercial considerations.
Quantitative performance
Target assessment time
4–6 h
vs. 3–4 months manual
Acceleration factor
~200×
end-to-end target review
Toxicity prediction sensitivity
83%
10 / 12 true positive
Toxicity prediction specificity
78%
7 / 9 true negative
Cancer antigens indexed
>10,000
unified knowledge graph
PubMed abstracts integrated
50 M
semantic retrieval
Keywords
CAR-T cell therapyMulti-agent AILarge language modelsTarget discoveryToxicity predictionPatent intelligencePrecision oncologyAutonomous drug discovery
How to cite
Ni Y, Zhu L, Li S. Bio AI Agent: A multi-agent artificial intelligence system for autonomous CAR-T cell therapy development with integrated target discovery, toxicity prediction, and rational molecular design. arXiv:2511.08649 [q-bio.QM]; 2025.
Comments