Research

Open research from LiteFold: LiteMol-1, AminoWeb and BenchPLM, covering molecular foundation models, protein datasets and protein language model benchmarks.

• LiteMol-1: A Multi Molecule Foundation Diffusion Language Model for Agents: LiteFold's first foundation model for agent-driven molecular design, generating small molecules, peptides, macrocycles, and PROTACs.
• BenchPLM: Proteins Aren't Sentences: Why Bigger Protein Models Don't Win: A frozen-embedding benchmark across protein language models — ESM-C, DPLM, ESM-2, and ProGen — probing downstream tasks, evolutionary geometry, context usage, and sparse-data transfer to ask which representations actually carry biological signal.
• AminoWeb: Crystallizing the Web for the Finest Protein Data at Scale: Open-sourcing AminoWeb — a 29-dataset, ~7.56 TB Parquet-backed collection of cleaned, standardized, ML-ready protein data spanning sequences, structures, function, evolution, and experimental assays, curated around real biological constraints.