Hill-Climbing Biology

Models, environments and datasets that let a molecule climb against everything it has to satisfy at once. Affinity is the easy objective. A molecule that reaches the clinic has to be selective, stable, soluble, synthesizable and safe, all at the same time.

Three pillars, one design loop

• Models: generative models that propose molecules conditioned on a target, a pocket and a full property profile, not on binding alone.
• Environments: scoring environments a sampler can query mid-generation, so design becomes a loop that closes rather than a single forward pass.
• Datasets: curated, ML-ready protein data at scale across 29 datasets and 7.56 TB, covering sequences, structures, function, evolution and assays.

LiteMol-1

A multi-molecule foundation diffusion language model for agents, generating across small molecules, linear and cyclic peptides, macrocycles and PROTACs. An agent can steer it at sampling time, searching the design space under many objectives at once.

Modalities

• Proteins and peptides: de novo binders, cyclic and stapled peptides, non-canonical residues.
• Small molecules: pocket-conditioned design, PROTACs and glues, ADMET-aware optimization.