ARTFEED — Contemporary Art Intelligence

LLMs Predict Scalability of Metal-Organic Framework Syntheses

ai-technology · 2026-04-25

Researchers have developed ESU-MOF, a dataset and machine learning strategy that fine-tunes large language models to predict the scalability potential of metal-organic framework (MOF) syntheses with 91.4% accuracy. The work addresses the bottleneck between MOF discovery and industrial deployment by mining fragmented literature. The positive-unlabeled learning approach enables rapid data-driven triage for industrial MOF discovery. The study is published on arXiv (ID 2604.20899) under the Condensed Matter > Materials Science category.

Key facts

  • ESU-MOF is a literature-mined dataset for MOF scalability.
  • Fine-tuned large language models achieve 91.4% accuracy.
  • Positive-unlabeled learning strategy is used.
  • Scalable synthesis is a gate between MOF discovery and industrial deployment.
  • Scale-up know-how is fragmented across reports.
  • The work enables data-driven triage for industrial MOF discovery.
  • Published on arXiv with ID 2604.20899.
  • Categorized under Condensed Matter > Materials Science.

Entities

Institutions

  • arXiv

Sources