cs.AISep 22, 2026

ChatT2: An Adaptive Framework for Developing a Large Language Model-Based Agent for Natural Product Domain Research

Authors: Yihan Wang, Qiandi Gao, Yihui Zhuang, Liangjun Ge, Heqian Zhang, Jiaquan Huang, Zhiwei Qin

Organizations: Center for Biological Science and Technology, Advanced Institute of Natural Sciences, Beijing Normal University, Zhuhai, Guangdong, 519087, China.

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Scientific investigations into microbial natural products (NPs) present significant challenges for novices, largely due to the complexity of microbial systems, biochemical diversity, technical skill requirements, and the demands of bioinformatics and data analysis processes. To address these issues, we introduce ChatT2, a large language model (LLM)-based agent that is specifically tailored to the unique characteristics of bacterial type II polyketides. These polyketides form a structurally distinct and therapeutically important NP family. ChatT2 was developed within an autonomous multiagent framework composed of a mentor, an executor, and an evaluator, each with defined responsibilities. The mentor acts as an intermediary between ChatT2 and the user, utilizing chain-of-thought prompting to refine the intent of the user. Under the guidance of the mentor, the executor synthesizes multimodal information via retrieval-augmented generation techniques and seamlessly integrates bioinformatics and cheminformatics tools. The evaluator ultimately assesses the output of the executor to ensure the richness and accuracy of the retrieved information. Our research highlights how ChatT2, designed with this multiagent framework, addresses the challenges faced by general LLMs in terms of understanding limited, specialized corpora and complex biological information and provides both experts and novices with a valuable tool for exploring various NPs of interest. The ChatT2 webserver can be accessed at https://chatt2.site/#/chat.

Figures & tables

Figure 1 : The workflow of ChatT2 framework. Core process: The Mentor and Executor engage in t rounds of interaction. In each round, the Mentor submits a query to the Executor, who processes it and retrieves relevant contextual information from various knowledge bases. This information may include text, images, citations, and other supporting materials to address the current query. Based on the user’s original intent and the results returned by the Executor in previous rounds, the Mentor formulates the next query. After completing t rounds, the Mentor summarizes all the questions and answers from the interaction into a final response, which is then returned to the user. Evaluation process: The Evaluator interacts with the Executor before, during, and after the retrieval and answer generation process to enhance response quality. Raise error: If the Evaluator identifies anomalies during the Executor’s retrieval process, it reports them to the Mentor.
Figure 2 : Multiple retrieval source of executor. When the Executor receives a query, it transforms the query into four different types of searches across a relational database, vector database, image database, and tool repository to retrieve relevant information. Relational and Vector Databases: The query is transformed into SQL statements and text embeddings: SQL statements are executed to retrieve table information related to the query, presented within <Table> tags. During this process, relevant literature sources are gathered, forming a collection of papers denoted as IsI_{s} . Text embeddings are used to perform hierarchical semantic similarity matching based on cosine similarity. First, a semantically similar set of papers is identified, denoted as IvI_{v} . The embeddings of the query are then used to match the semantic similarity of document chunks from Iv∪IsI_{v}\cup I_{s} , extracting text relevant to the query from the body of the papers. This information is presented within <Text> tags. Image Database: The query is matched against image titles using BM25 scores. High-scoring images are identified, and their URLs are saved as image information, presented within <Figure> tags. Tool Repository: The Executor parses key information from the query as input for prepackaged executable scripts. The execution results of these scripts are treated as tool repository information, encapsulated within <Tool> tags.
Figure 3 : Comparison results of responses from different models and ChatT2. (a) The figure shows the comparison of the response texts between the original LLM and ChatT2, which uses the same LLM as its brain, for the same question; (b) Radar chart comparing the performance of responses from ChatT2 and different models across five evaluation metrics. Violin plot illustrating the rank score distributions of responses from ChatT2 and different models.
AssistantQuestion type
BiosynthesisStructureFunctionApplicationClassification
ChatT2 (GPT-4o mini)0.8440.7120.8080.8040.624
GPT-4o mini0.0440.0520.0520.0720.164
Claude-3.5-Sonnet00.01600.0080
Llama 3.1-405B0.1120.220.140.1160.212
Mean spearman correlation (r=5)0.7140.6230.6210.6470.607
Table 1 : Winning rate of different models and spearman correlation across different types of questions.
Figure 4 : The overall result of ChatT2 with CoT and without CoT between the Mentor and Executor. (a) The panel shows the questions formulated by the Mentor over multiple interaction rounds with the Executor; (b) After applying CoT techniques, the evaluation results for the answers are presented in terms of the cumulative number of explored documents, word richness, and text length per round.
Figure 5 : An open-ended question-answer demonstration operating in Auto mode.

Explore similar work

CardsList
Jul 16, 2026cs.AI

RetroAgent: Harnessing LLMs to Search Over Structured Memory for Agentic Retrosynthesis Planning

Multi-step retrosynthesis planning seeks to decompose a target molecule into commercially available building blocks through a sequence of feasible reactions. The vast combinatorial search space makes this task challenging even for expert chemists. Traditional methods combine tree search with offline-trained value networks that score candidates in isolation, without reasoning about complete multi-step routes. Recent work leverages Large Language Models (LLMs) for this task, but relies on simple interfaces that limit exploration of the full search space. We introduce RetroAgent, an LLM agent that bridges symbolic search and neural reasoning through a harness with structured memory. Through memory and chemistry tools, the agent observes the full search state, including explored routes, available alternatives, and properties of intermediates, enabling informed decisions grounded in both global progress and domain knowledge. Experiments on in-distribution and out-of-distribution benchmarks demonstrate that RetroAgent delivers strong performance and generalization.

Yanqiao Zhu, Jingru Gan, Xiaoqi Sun +8

UCLA · MIT · Amazon +1

Large Language Model PlanningAgentic MemoryAgenticPeak Memory
AskPDF
Aug 6, 2026cs.AI

Research Assistant: AstraZeneca's Agentic System for R&D

We describe Research Assistant, an internal LLM-based system developed at AstraZeneca to help scientists and clinicians explore biomedical questions across a broad range of data sources. The system provides a chat-style interface that brings together evidence from scientific literature, knowledge graphs, chemistry, clinical trials, safety resources, expression data, and internal experimental systems. It supports both a fast mode for direct question answering and a multi-step mode for more complex research tasks. Responses are grounded in retrieved evidence and linked back to the original sources, allowing users to review and further explore the underlying data. In this technical note, we outline the system architecture, the main design choices behind the product, and lessons learned from deploying it at scale to support day-to-day R&D workflows across AstraZeneca.

Piotr Grabowski, Mohamed Alameen, Jorge Bretones +16

AstraZeneca, RDIT, Biological Insights Knowledge Graph

Research AutomationBiomedical TextExpression
AskPDF
Jun 8, 2026physics.chem-ph

My Chemical Harness: Evolutionary Molecular Design over Synthetic Pathways with Large Language Model Agents

Designing molecules with target properties is most useful when candidate structures are accompanied by feasible synthetic routes. We introduce My Chemical Harness, a route-native evolutionary framework for goal-directed molecular design in which the search population consists of executable synthetic pathways rather than isolated molecular graphs. Each route is built from purchasable building blocks and reaction templates, executed by deterministic chemistry tools, and scored through task-specific molecular oracles. Large language models (LLMs) are used only as strategy controllers that select high-level preferences over route length, move type, reaction families, motifs, and exploration pressure, while local code performs route construction, validation, deduplication, scoring, selection, and memory updates. This separation lets the LLM guide exploration without allowing it to introduce hallucinated products or unsupported reaction steps. On a soluble epoxide hydrolase proxy task, our LLM agent improves over single pass LLM and deterministic controllers, reaching state-of-the-art performance across the sEH score, synthetic accessibility score, and AiZynthFinder success rate metrics. These results suggest that constrained LLM agents can play a significant role in molecular discovery without requiring training, fine-tuning, or dedicated generative models.

César Ojeda, Darius A. Faroughy, Maryam Karimi +3

Institute of Mathematics, Faculty of Science, University of Potsdam, Campus Golm, Building 9, Karl-Liebknecht-Str. 24–25, 14476 Potsdam-Golm, Germany · NHETC, Department of Physics and Astronomy, Rutgers University, Piscataway, NJ 08854, USA · Potsdam Transfer, University of Potsdam, Karl-Liebknecht-Str. 24-25, 14476 Potsdam. +1

Molecular Design
AskPDF