LLM-Fusion introduces a novel approach that leverages large language models to enhance material discovery by intelligently combining multiple types of material data.

• Integrates molecular representations (SMILES/SELFIES) with textual descriptions for more accurate property prediction
• Outperforms traditional single-modality and simple fusion methods
• Uses specialized cross-attention mechanisms to create richer material representations
• Demonstrates practical applications in accelerating the discovery of materials with targeted properties

This engineering breakthrough offers significant potential for reducing the time and cost of developing new materials across industries, from electronics to medicine, by enabling more efficient exploration of vast material spaces.

LLM-Fusion: A Novel Multimodal Fusion Model for Accelerated Material Discovery