Our Datasets

Synthetic datasets designed for training state-of-the-art protein structure prediction models. Supercharge your models and go beyond the PDB.

Who's already using synthetic datasets?

Here are the world-leading models that already use synthetic data in their training curriculum.

Foundational co-folding models

Demonstrated significant performance advances by training on large-scale self-distillation datasets, especially for biologics/antibodies.

Genesis Molecular AI

Technical report extensively mentions the data-scaling benefits observed when training on PDB-based synthetic data.

Boltz.bio

Boltz2 pre-print mentions 3 separate molecular dynamics datasets used in model training.

Large-scale model inferences with confidence filtering for high-quality synthetic structures

→Base Models: any open-source model (OpenFold3p, Boltz1/2), including your custom models
→Confidence Threshold: calibrated pLDDT
→Sequence: clustered to avoid redundancy and maximize diversity
→Scale: Millions of structures

Extended molecular dynamics simulations capturing protein dynamics and conformational states

Join our waitlist for early access to pre-made packages and custom generation