MIT multimodal model links 3D crystal structures with language to improve materials discovery
MIT researchers published a Nature Communications paper describing a multimodal AI model that combines 3D atomic crystal structures with natural‑language descriptions via contrastive learning to produce unified textual narratives of materials. The model outperformed a text‑only baseline at identifying candidate materials (including promising solid‑state lithium‑ion electrolyte candidates) by better aligning geometric structure and functional descriptions.