La-Proteina Now Available on Vecura for Atomistic Protein Design

What is La-Proteina?

La-Proteina is a deep generative model that simultaneously produces full-atom three-dimensional protein structures and co-designed amino acid sequences using a novel partially latent flow matching approach. Unlike backbone-only models, it explicitly represents the alpha-carbon backbone while encoding sequence and side-chain details in per-residue latent variables, producing output directly in all-atom PDB format. It is especially useful for de novo protein design and atomistic motif scaffolding, where functional structural elements must be embedded in stable protein scaffolds.

What can users do with La-Proteina on Vecura?

With La-Proteina on Vecura, users can:

• Generate novel full-atom protein structures up to 800 residues with co-designed sequences

• Perform atomistic motif scaffolding to embed functional binding sites or catalytic motifs in stable protein contexts

• Produce diverse, designable protein backbones ready for wet-lab expression or computational refinement

• Generate multiple independent samples per request with configurable parameters for reproducibility

What the output means

The output provides generated full-atom protein structures in PDB format and corresponding co-designed amino acid sequences. These structures include backbone atoms (N, CA, C, O), beta-carbons, and complete side-chain atoms with chain IDs, ready for visualization and downstream tools like ProteinMPNN or ESMFold for re-scoring.

This output should be used to support scientific decision making. It does not replace experimental validation.

Why this matters

Protein design requires simultaneously optimizing sequence and structure to create stable, functional molecules. Traditional approaches separate backbone generation from side-chain packing, often losing critical atomic-level details. La-Proteina's partially latent flow matching approach maintains explicit backbone geometry while capturing complex side-chain interactions in a flexible latent space, achieving over 75% co-designability—doubling the success rate of prior methods.

This capability enables new applications in therapeutic enzyme design, binder development, and functional protein engineering. By generating diverse, viable protein candidates at scale, researchers can accelerate the discovery of novel proteins for drug development, industrial biocatalysis, and synthetic biology.

• Developed by: NVIDIA BioNeMo

• Source: Official documentation and model card

• Reference: Original paper (arXiv:2507.09466) | GitHub repository | Project page