Researchers have developed MIST, a new family of molecular foundation models designed to accelerate discovery and exploration within chemical space. These models, trained on extensive unlabeled datasets using a novel tokenizer called Smirk, demonstrate strong performance across over 400 structure-property relationships. MIST has shown success in predicting scent profiles, a task not explicitly part of its training, and has been applied to real-world problems in areas like electrolyte screening and organometallic stereochemistry. AI
Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →
IMPACT Accelerates materials discovery and design by providing scalable navigation of chemical space.
RANK_REASON Academic paper detailing a new family of foundation models for chemical discovery.