Predicting Scale-Up of Metal-Organic Framework Syntheses with Large Language Models

2025-04-14 · Source: cs.AI updates on arXiv.org · Field: Science & Research — Physical Sciences & Chemistry, Mathematics & Computational Sciences, Research Methodology & Innovation · Depth: Expert, extended

Summary

ESU-MOF is a novel literature-mined dataset and learning strategy that fine-tunes large language models (LLMs) to predict the scalability of Metal-Organic Framework (MOF) syntheses. This approach addresses the challenge of fragmented scale-up knowledge. The workflow uses an LLM agent to extract 3,568 synthesis protocols from scientific literature, creating the ESU-MOF dataset. A separate LLM, fine-tuned with a positive-unlabeled (PU) learning strategy, achieves 91.4% balanced accuracy on a gold benchmark and a 94.5% ROC-AUC on a deployment benchmark. This model significantly outperforms traditional machine learning and untuned LLM baselines, demonstrating that latent information in small-scale synthesis descriptions can predict gram-scale or larger production potential.

Key takeaway

For research scientists evaluating new MOF discoveries, this LLM-driven workflow offers a critical tool to predict scale-up potential early. You can rapidly triage synthesis protocols, identifying those with high manufacturability for gram-scale or larger production. This enables more informed prioritization of experimental efforts and industrial engagement, potentially accelerating the path from lab discovery to commercial realization.

Key insights

LLMs fine-tuned with PU learning can predict MOF synthesis scalability from small-scale literature data.

Principles

• Scalability is a protocol property, not intrinsic to the MOF.
• Literature contains latent scale-up knowledge.
• PU learning is suitable for incomplete positive labels.

Method

An LLM extracts structured synthesis data, which forms the ESU-MOF dataset. A second LLM is fine-tuned using positive-unlabeled learning, followed by PU correction and Platt scaling for calibrated scalability scores.

In practice

• Use LLMs for data mining chemical synthesis parameters.
• Apply PU learning when true negatives are scarce.
• Prioritize MOF protocols with benign solvents and moderate conditions.

Topics

• Metal-Organic Frameworks
• Large Language Models
• Scalable Synthesis
• Positive-Unlabeled Learning
• Chemical Data Mining
• Materials Discovery

Code references

• zzhenglab/MOF-Scaleup

Best for: AI Scientist, Research Scientist, Domain Expert

Related on AIssential

• RxnNano:Training Compact LLMs for Chemical Reaction and Retrosynthesis Prediction via Hierarchical Curriculum Learning — Deep chemical understanding and structured training enable compact LLMs to outperform larger models in reaction prediction.
• Large language model agents accelerate inverse design of metal-organic frameworks for gas separation — LEMO Agent demonstrates large language model agents can accelerate inverse design of complex materials like MOFs for gas separation.
• LEAP: A closed-loop framework for perovskite precursor additive discovery — Coupling a domain-specialized LLM with active learning and expert review accelerates perovskite additive discovery under low-data conditions.
• Coupling Language Models with Physics-based Simulation for Synthesis of Inorganic Materials — Coupling LLMs with physics-based simulations offers more viable inorganic material synthesis routes than classical algorithms, leveraging LLMs' implicit priors.
• Can Agents Price a Reaction? Evaluating LLMs on Chemical Cost Reasoning — LLM agents struggle with real-world chemical procurement cost estimation, even with tool access, due to parsing and reasoning failures.
• OLEDLM: A Unified Language Model for OLED Molecular Design — OLEDLM uses a multi-stage language model to generate OLED molecules directly from desired optoelectronic properties.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.