NVIDIA-BioNeMo/Proteina-Complexa

What it solves

Proteina-Complexa addresses the challenge of atomistic protein binder design. It unifies two previously distinct approaches—conditional generative modeling and sequence optimization (hallucination)—into a single framework to create novel proteins that bind to specific targets, including other proteins or small-molecule ligands.

How it works

The system uses a flow-matching generative model that jointly models backbone geometry, side-chain conformations, and sequences. It was pretrained on "Teddymer," a large-scale synthetic dataset of binder-target pairs, and further refined with high-quality experimental multimers. During inference, it employs test-time optimization using reward models (such as AlphaFold2, RoseTTAFold3, and various force fields) to refine the generated binders.

Who it’s for

This tool is designed for computational biologists, drug discovery researchers, and protein engineers who need to design high-affinity binders for therapeutic or industrial applications.

Highlights

• Unified Framework: Combines generative priors with search-based optimization to outperform previous hallucination and generative methods.
• Versatile Design: Supports protein-protein binder design, protein-ligand binder design, and motif scaffolding (AME).
• Integrated Pipeline: Includes built-in sequence design (via ProteinMPNN and others) and structure prediction validation (via AlphaFold2, ESMFold, and RoseTTAFold3).
• Experimental Validation: Designs have been wet-lab validated to demonstrate real binding activity.