Ilmu Komputer & AI editorial
Open AccessOA2026
ChatT2: An Adaptive Framework for Developing a Large Language Model-Based Agent for Natural Product Domain Research
A multiagent LLM system for bacterial type II polyketide research
Yihan Wang; Qiandi Gao; Yihui Zhuang; Liangjun Ge; Heqian Zhang; Jiaquan Huang; Zhiwei Qin· 2026· DOI 10.48550/arXiv.2609.25620
The core problem
Microbial natural products (NPs) are a rich source of therapeutic agents, but their investigation is challenging for novices due to the complexity of microbial systems, biochemical diversity, technical skill requirements, and the demands of bioinformatics and data analysis. General large language models (LLMs) struggle with limited, specialized corpora and complex biological information. To address these issues, the authors introduce ChatT2, an LLM-based agent specifically tailored to bacterial type II polyketides, a structurally distinct and therapeutically important NP family. ChatT2 is developed within an autonomous multiagent framework composed of a mentor, an executor, and an evaluator, each with defined responsibilities. The webserver is available at https://chatt2.site/#/chat.
Innovation
The authors report that ChatT2, designed with this multiagent framework, addresses the challenges faced by general LLMs in understanding limited, specialized corpora and complex biological information. It provides both experts and novices with a valuable tool for exploring various NPs of interest. The system was specifically tailored to bacterial type II polyketides, demonstrating its ability to handle a structurally distinct and therapeutically important NP family. The webserver is accessible at https://chatt2.site/#/chat.
Microbial natural products (NPs) are a rich source of therapeutic agents, but their investigation is challenging for novices due to the complexity of microbial systems, biochemical diversity, technical skill requirements, and the demands of bioinformatics and data analysis. General large language models (LLMs) struggle with limited, specialized corpora and complex biological information. To address these issues, the authors introduce ChatT2, an LLM-based agent specifically tailored to bacterial type II polyketides, a structurally distinct and therapeutically important NP family. ChatT2 is developed within an autonomous multiagent framework composed of a mentor, an executor, and an evaluator, each with defined responsibilities. The webserver is available at https://chatt2.site/#/chat.
ChatT2 employs a multiagent architecture with three core components: a mentor, an executor, and an evaluator. The mentor acts as an intermediary between ChatT2 and the user, utilizing chain-of-thought prompting to refine the user's intent. Under the mentor's guidance, the executor synthesizes multimodal information via retrieval-augmented generation (RAG) techniques and seamlessly integrates bioinformatics and cheminformatics tools. The evaluator assesses the executor's output to ensure the richness and accuracy of the retrieved information. This design enables the system to handle the complexities of bacterial type II polyketide research.
Why it matters
The multiagent framework of ChatT2 effectively mitigates the limitations of general LLMs in specialized domains. By decomposing tasks among mentor, executor, and evaluator, the system ensures accurate intent refinement, comprehensive information retrieval, and rigorous output validation. The integration of RAG with bioinformatics and cheminformatics tools enables seamless synthesis of multimodal information. This approach not only aids novices in overcoming technical barriers but also provides experts with a powerful research assistant. Future work may extend the framework to other NP families and incorporate additional tools. The success of ChatT2 highlights the potential of multiagent LLM systems in advancing domain-specific scientific research.
Who should read this
CS practitioners and researchers
Opening member content…