MolLean is now online, introducing a process-verifiable language for chemical reasoning with a Lean-based kernel and interactive verification. Read the paper ↗
AI × Molecular Science
Jiatong Li
Building AI co-scientists for molecular discovery.
I work on molecular representations, chemical reasoning, and tool-using agents. My goal is to build AI co-scientists that help researchers explore and test molecular ideas.
Latest updates
News
STAMP is now online, presenting a topology-aware molecular language for fragment-level learning and precise structural editing. Read the paper ↗
I am serving as a Teaching Assistant for COMP5584, Large Language Models for Agents, at The Hong Kong Polytechnic University this semester.
I will return to NUS as an incoming Research Fellow with the BlueWhale Lab ↗, following my Visiting Scholar appointment from January to July 2026.
Earlier notes (8)Close earlier notes
Our perspective on molecular LLM agents is online. Molecular LLM Agents: From Architectural Design to Scientific Autonomy ↗
Our chemical chain-of-thought study is online. Chemical Chain-of-Thought Functions as a Hallucination-Prone Molecular Scratchpad ↗
Our analysis of molecular LLM generalization is online. Do LLMs Truly Generalize in the Molecular Domain? ↗
Our medical reasoning survey is accepted by Machine Intelligence Research. Aligning Clinical Needs and AI Capabilities ↗
Two papers are accepted by KDD 2026: Speak-to-Structure ↗ and Chem-R ↗, my first work as a co-corresponding author.
MolViBench is released, introducing a benchmark for molecular vibe coding across real-world drug-discovery workflows. Code ↗
MolReFlect is accepted by IEEE TKDE. Code ↗
CMPhysBench is accepted by ICLR 2026. Paper ↗
AI co-scientist research agenda
Research directions
My research connects molecular representations, chemical reasoning, and scientific tool use. These four questions guide the work.
FIELD NOTE / 01 Understand molecules Representations & molecular state Explore this direction FIELD NOTE / 02 Reason with evidence Chemical reasoning & verification Explore this direction FIELD NOTE / 03 Work through tools Scientific agents & workflows Explore this direction FIELD NOTE / 04 Test the limits Generalization & evaluation Explore this direction
01 / The research question
How can an AI maintain an explicit molecular state?
I study representations that connect molecular structures, fragments, and language. These representations let a model describe a molecule and specify structural edits.
Start with an explicit molecular state.02 / The research question
How can we verify each step of chemical reasoning?
I develop methods for checking intermediate molecular states and chemical operations. I also study how chain-of-thought explanations relate to a model’s predictions.
Verify the intermediate steps.03 / The research question
How can agents carry out molecular research tasks?
I study agents that translate molecular design goals into tool calls and code. My evaluations test the resulting programs on drug-discovery workflows.
Connect research questions to tools and code.04 / The research question
How do models behave on unfamiliar scientific problems?
I test models with molecular graph edits and problems from physics and medicine to understand where their predictions change or fail.
Test beyond familiar examples.Each direction links to the papers behind it.
From the notebook · 2026
Latest research
A sketch of each idea, and the work behind it. Browse the collection to explore molecular languages, scientific reasoning, and agents.
arXiv · Oct 202601
MolLean
Chemical reasoning, checked step by step.
Inside this paper
MolLean: A Process-Verifiable Language for Chemical Reasoning
A state-grounded chemical reasoning language whose Lean-based kernel checks molecular transitions and assertions, with whole-trace and interactive feedback.
arXiv · Oct 202602
STAMP
A molecular language built from reusable fragments.
Inside this paper
STAMP as a topology-aware molecular language for fragment-level learning and editing
A fragment-tree molecular language that makes motifs, attachment sites, and topology explicit for representation learning and executable editing.
arXiv · Aug 202603
Molecular LLM Agents
From tool-using assistants to scientific autonomy.
Inside this paper
Molecular LLM Agents: From Architectural Design to Scientific Autonomy
A unified architecture for molecular LLM agents and a four-level autonomy ladder spanning assistive workflows to scientific-agenda agents.
arXiv · Jul 202604
Chemical CoT
Tracing the role of chemical scratchpads.
Inside this paper
Chemical Chain-of-Thought Functions as a Hallucination-Prone Molecular Scratchpad
Chemical CoT can be causally useful as a structural scratchpad while remaining an unreliable explanation of how an answer was reached.
arXiv · Jul 202605
Molecular Generalization
How far does molecular understanding generalize?
Inside this paper
Do LLMs Truly Generalize in the Molecular Domain? A Perturbation-Based Analysis
Controlled graph edits reveal a narrow local trust region in molecular LLMs and show how in-context tuning can partially stabilize it.
Machine Intelligence Research · 202606
Medical Reasoning
Connecting clinical needs with model capabilities.
Inside this paper
Aligning Clinical Needs and AI Capabilities: A Survey on LLMs for Medical Reasoning
A survey of medical reasoning from clinical and computational perspectives, with a five-level benchmark evaluating 18 models.
arXiv · May 202607
MolViBench
Can generated code do useful molecular science?
Inside this paper
MolViBench: Evaluating LLMs on Molecular Vibe Coding
A benchmark for executable molecular coding across 12 real-world drug-discovery workflows.
KDD · 202608
Speak-to-Structure
One design brief. Many possible molecules.
Inside this paper
Speak-to-Structure: Open-Domain Natural Language-Driven Molecule Generation
S²-Bench evaluates one-to-many molecular design from natural-language requirements.
KDD · 202609
Chem-R
Learning to reason like a chemist.
Inside this paper
Chem-R: Learning to Reason as a Chemist
A three-phase chemical reasoning model spanning molecular- and reaction-level tasks.
IEEE TKDE · 202610
MolReFlect
Aligning molecular details with their descriptions.
Inside this paper
MolReFlect: Fine-Grained Alignments Between Molecules and Texts
An in-context reflection framework for precise molecule–text alignment.
ICLR · 202611
CMPhysBench
Putting calculation-intensive reasoning to the test.
Inside this paper
CMPhysBench: Evaluating LLMs in Condensed Matter Physics
A fine-grained benchmark for calculation-intensive reasoning in condensed matter physics.
Selected
Recent publications
Recent papers on molecular representations, chemical reasoning, and scientific agents.
- 2026arXiv
MolLean: A Process-Verifiable Language for Chemical Reasoning
Jiatong Li, Yatao Bian
- 2026arXiv
STAMP as a topology-aware molecular language for fragment-level learning and editing
Yuxuan Ren, Jiatong Li, Ziqiao Meng, Yatao Bian
- 2026arXiv
Molecular LLM Agents: From Architectural Design to Scientific Autonomy
Jiatong Li, Wengyu Zhang, Weida Wang, Yuxuan Ren, Wei Liu, Chenyang Mao, Yuqiang Li, Yatao Bian, Changmeng Zheng, Xiaoyong Wei, Qing Li
- 2026arXiv
Chemical Chain-of-Thought Functions as a Hallucination-Prone Molecular Scratchpad
Jiatong Li, Yuxuan Ren, Weida Wang, Xiaoyong Wei, Yatao Bian
- 2026arXiv
Do LLMs Truly Generalize in the Molecular Domain? A Perturbation-Based Analysis
Jiatong Li, Weida Wang, Changmeng Zheng, Shufei Zhang, Yatao Bian, Xiao-yong Wei, Qing Li
- 2026MIR
Aligning Clinical Needs and AI Capabilities: A Survey on LLMs for Medical Reasoning
Qi Peng*, Jiatong Li*, Sirui Huang, Yiyang Jiang, Kaisong Gong, et al.
- 2026arXiv
MolViBench: Evaluating LLMs on Molecular Vibe Coding
Jiatong Li, Yuxuan Ren, Weida Wang, Changmeng Zheng, Xiao-yong Wei, Qing Li, Yatao Bian
- 2026IEEE TKDE
MolReFlect: Towards In-Context Fine-grained Alignments between Molecules and Texts
Jiatong Li, Yunqing Liu, Wei Liu, Jingdi Lei, Di Zhang, Wenqi Fan, Dongzhan Zhou, Yuqiang Li, Qing Li
- 2026KDD
Speak-to-Structure: Evaluating LLMs in Open-domain Natural Language-Driven Molecule Generation
Jiatong Li, Junxian Li, Weida Wang, Yunqing Liu, Changmeng Zheng, Yatao Bian, et al.
- 2026KDD
Chem-R: Learning to Reason as a Chemist
Weida Wang, Benteng Chen, Di Zhang, et al., Jiatong Li, et al. · Co-corresponding author
- 2026ICLR
CMPhysBench: A Benchmark for Evaluating Large Language Models in Condensed Matter Physics
Weida Wang*, Dongchen Huang*, Jiatong Li*, Tengchao Yang, et al. · Co-first author
- 2026CVPR
IAG: Input-Aware Backdoor Attack on VLM-Based Visual Grounding
Junxian Li, Beining Xu, Simin Chen, Jiatong Li, Jingdi Lei, Haodong Zhao, Di Zhang
- 2025IEEE TKDE
Large Language Models are In-Context Molecule Learners
Jiatong Li, Wei Liu, Zhihao Ding, Wenqi Fan, Yuqiang Li, Qing Li
- 2025arXiv
Mol-R1: Towards Explicit Long-CoT Reasoning in Molecule Discovery
Jiatong Li, Weida Wang, Qinggang Zhang, Junxian Li, Di Zhang, Changmeng Zheng, et al.
Background
Research journey
Training across information security, natural language processing, and AI-driven molecular discovery.
Education
- 2023–Present Ph.D. Candidate, Computing
The Hong Kong Polytechnic University · HKPFS
- 2021–2022 Master of Information Technology
The University of Melbourne · Dean’s Honor List
- 2017–2021 B.Eng. in Information Security
Shanghai Jiao Tong University
Research experience
- Incoming Research Fellow
BlueWhale Lab ↗ · National University of Singapore
- 2026.01–07 Visiting Scholar
BlueWhale Lab ↗ · National University of Singapore
- 2023–2024 Algorithm Researcher
Shanghai AI Laboratory
- 2021–2022 Algorithm Researcher
Noah’s Ark Lab
Beyond papers
Teaching & service
Teaching Assistant · PolyU
- COMP5584 Large Language Models for Agents · 2026/27 Semester 1
- COMP2411 Database Systems
- COMP4434 Big Data Analytics
- COMP6703 Advanced Topics in Data Analytics
- COMP6713 Advanced Large Language Models and Beyond
Program committees & reviewing
PC: AAAI 2024–2026, IIAI AAI 2026, PRICAI 2024–2025, WISE 2024.
Reviewer: ICLR, NeurIPS, ICML, ACL, KDD, AISTATS, IEEE TKDE, Journal of Cheminformatics, npj Digital Medicine, and more.
Tutorials & workshops
Recognition
Honors
HKSAR Government
The University of Melbourne
Mathematical Contest in Modeling
On repeat
A little life beyond the lab.
A song for the hours between molecules and models.
Collaborate
Interested in AI for molecular science?
I welcome conversations about chemical reasoning, scientific agents, and research collaborations.
Live readership
Most-clicked papers
- Gathering the latest clicks…
Paper-link clicks · Scroll for the full ranking.
Live homepage visits
Where visitors arrive from
North America0 South America0 Europe0 Africa0 Asia0 Oceania0
- Loading visitor locations…
Homepage visits by country/region. Paper clicks have their own count.