From Literature to Experimental Conditions: Large Language Models for Co-Crystal Synthesis Design

Nina Gubina, Bagramian Anita, Kadochnikova Margarita Sergeevna, Nikita A. Vasilev, Ivan Yu. Chernyshov, Andrei Dmitrenko

Track: Track 1: Original Research/Position/Education/Attention Track

Keywords: co-crystal synthesis, pharmaceutical co-crystals, information extraction, scientific literature mining, structured synthesis database, synthesis condition prediction, multi-agent systems, applied artificial intelligence

TL;DR: CoSyn is an LLM-based agentic system that turns scattered co-crystal synthesis literature into a structured database and actionable synthesis-condition recommendations.

Abstract: Co-crystallization enables modulation of drug physicochemical properties, yet practical synthesis design remains largely empirical. Existing AI methods mainly address molecular pair selection, leaving synthesis-condition design underexplored because the required experimental knowledge is scattered across scientific articles. We present CoSyn, a framework that uses large language models to convert dispersed procedure descriptions into machine-readable synthesis records, enabling experimental condition recommendations for new molecular pairs and tool-augmented assistance. Using manually annotated datasets, we evaluate both individual components and system-level performance. Our results highlight the potential of a modular LLM-based approach to organize literature-derived evidence and support co-crystal synthesis design beyond pair-level screening.

Submission Number: 144