LLM-AutoSciLab: Closed-Loop Scientific Discovery via Active Experimentation with LLMs

Abstract

Scientific discovery is a closed-loop process in which hypotheses guide data acquisition and observations refine the hypothesis space. Yet most approaches reduce discovery to supervised learning over fixed datasets, where limited observations can support multiple plausible mechanisms that fit locally but fail to generalize. Thus, the key challenge is selecting informative observations to resolve uncertainty, shifting the focus from static inference to adaptive data acquisition. To address this, we propose LLM-AutoSciLab, a closed-loop framework that couples hypothesis generation with hypothesis-conditioned experiment selection and mechanism refinement. Rather than fitting models to passively collected data, LLM-AutoSciLab iteratively proposes plausible hypotheses, selects informative experiments to distinguish or refine them, and updates its state using the resulting evidence. To evaluate dynamic, closed-loop scientific discovery with active data acquisition, we introduce ActiveSciBench, comprising two datasets: ActiveSciBench-Chem with 57 enzyme-kinetics tasks and ActiveSciBench-GRN with 45 gene-regulatory-network tasks. These datasets model discovery as a budget-constrained process requiring adaptive experiment design, variable selection, and recovery of true mechanisms. Across NewtonBench, ActiveSciBench-Chem, and ActiveSciBench-GRN, LLM-AutoSciLab outperforms prior methods, achieving 67.6% and 35.1% symbolic accuracy on NewtonBench and ActiveSciBench-Chem, respectively, and 31.1% exact graph recovery on ActiveSciBench-GRN. Moreover, hypothesis-guided experimentation is 2-5x more sample-efficient than the strongest competing baselines. Code and data are available at: https://github.com/scientific-discovery/LLM-AutoSciLab

Beyond Scalar Objectives: Expert-Feedback-Driven Autonomous Experimentation for Scientific Discovery at the Nanoscale.
Ralph Bulanadi, Jefferey Baxter, Arpan Biswas +7
ACS nano1 Oct 2026Peer-reviewedCC BYFull textMaterials & polymersEurope PMC
Scalable Thermoset Informatics Integrated with Autonomous Experimentation for On-Demand Discovery
Marcus R Fratarcangeli, Ayush Jain, Farzad Gholami +4
Chemistry of materials : a publication of the American Chemical Society9 Oct 2026Peer-reviewedCC BYFull textMaterials & polymersEurope PMCEurope PMCPMC
Scientific World Models: Toward Closed-Loop Scientific Intelligence
Huan Wang, Zhenlin Zhao, Siming Dong
ChemRxiv29 Sep 2026PreprintCC BY-NCGeneral chemistryChemRxiv
Accelerating scientific discovery with Co-Scientist
Juraj Gottweis, Wei‐Hung Weng, Alexander Daryin +48
Nature19 May 2026Peer-reviewedLicence not statedFull text947 citedMedicinal chemistry & drug discoveryOpenAlex
Autonomous in-silico inorganic materials discovery via multi-agent physics-aware scientific reasoning
Alireza Ghafarollahi, Markus J. Buehler
npj Computational Materials8 Jul 2026Peer-reviewedCC BY-NC-ND 4.0Full text19 citedMaterials & polymersOpenAlexPMCEurope PMC

Cite

BibTeX RIS

BibTeX

@misc{kabra2026llmautoscilab,
  title = {{LLM-AutoSciLab: Closed-Loop Scientific Discovery via Active Experimentation with LLMs}},
  author = {Sanchit Kabra and Nikhil Abhyankar and Saaketh Desai and Prasad Iyer and Chandan K Reddy},
  howpublished = {arXiv (Cornell University)},
  year = {2026},
  month = may,
  doi = {10.48550/arxiv.2605.24043},
  url = {https://arxiv.org/abs/2605.24043},
  note = {CC BY},
}

RIS

TY  - UNPB
TI  - LLM-AutoSciLab: Closed-Loop Scientific Discovery via Active Experimentation with LLMs
AU  - Sanchit Kabra
AU  - Nikhil Abhyankar
AU  - Saaketh Desai
AU  - Prasad Iyer
AU  - Chandan K Reddy
PY  - 2026
DA  - 2026/05/21
PB  - arXiv (Cornell University)
DO  - 10.48550/arxiv.2605.24043
UR  - https://arxiv.org/abs/2605.24043
AB  - Scientific discovery is a closed-loop process in which hypotheses guide data acquisition and observations refine the hypothesis space. Yet most approaches reduce discovery to supervised learning over fixed datasets, where limited observations can support multiple plausible mechanisms that fit locally but fail to generalize. Thus, the key challenge is selecting informative observations to resolve uncertainty, shifting the focus from static inference to adaptive data acquisition. To address this, we propose LLM-AutoSciLab, a closed-loop framework that couples hypothesis generation with hypothesis-conditioned experiment selection and mechanism refinement. Rather than fitting models to passively collected data, LLM-AutoSciLab iteratively proposes plausible hypotheses, selects informative experiments to distinguish or refine them, and updates its state using the resulting evidence. To evaluate dynamic, closed-loop scientific discovery with active data acquisition, we introduce ActiveSciBench, comprising two datasets: ActiveSciBench-Chem with 57 enzyme-kinetics tasks and ActiveSciBench-GRN with 45 gene-regulatory-network tasks. These datasets model discovery as a budget-constrained process requiring adaptive experiment design, variable selection, and recovery of true mechanisms. Across NewtonBench, ActiveSciBench-Chem, and ActiveSciBench-GRN, LLM-AutoSciLab outperforms prior methods, achieving 67.6% and 35.1% symbolic accuracy on NewtonBench and ActiveSciBench-Chem, respectively, and 31.1% exact graph recovery on ActiveSciBench-GRN. Moreover, hypothesis-guided experimentation is 2-5x more sample-efficient than the strongest competing baselines. Code and data are available at: https://github.com/scientific-discovery/LLM-AutoSciLab
SN  - 2331-8422
ER  -

Abstract

Related

Cite