Designing Protein Binders Using the Generative Model Proteina-Complexa

2026-03-25 · Source: NVIDIA Technical Blog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computational Biology · Depth: Intermediate, medium

Summary

NVIDIA has released Proteina-Complexa, a generative model designed for de novo protein binder and enzyme design. This model utilizes a partially latent flow-matching framework, building on the La-Proteina model, to co-design both fully atomistic binder structures (backbone and side-chain) and their corresponding amino acid sequences. It was trained on over 1 million curated experimental and predicted structures from sources like the Protein Data Bank and AlphaFold. Proteina-Complexa integrates inference-time compute scaling with "reasoning" search algorithms to iteratively optimize designs, enhancing computational efficiency and binder quality. The model supports use cases including protein binders for disease-relevant protein targets, binders for small molecule targets in applications like targeted drug delivery, and de novo enzyme design for industrial biocatalysis. Extensive experimental validation, including testing 1 million candidates against 133 distinct protein targets, demonstrated its ability to generate high-affinity binders, even for challenging targets like sugar molecules.

Key takeaway

For AI Engineers and Research Scientists working on protein-based therapeutics or catalysts, Proteina-Complexa offers a unified platform to design high-affinity protein binders and enzymes. You should explore its co-design capabilities and inference-time optimization to generate novel biomolecules, potentially tackling targets previously considered intractable. Consider integrating its command-line interface into your workflows to accelerate experimental testing and discovery.

Key insights

Proteina-Complexa co-designs protein binders and enzymes using a latent flow-matching generative model with inference-time optimization.

Principles

• Co-designing structure and sequence improves binding affinity.
• Inference-time compute scaling refines generative model outputs.

Method

Proteina-Complexa employs a partially latent flow-matching framework to generate atomistic protein structures and sequences simultaneously, then refines designs using search algorithms and reward functions during inference.

In practice

• Design de novo protein binders for therapeutic targets.
• Create proteins binding specific small molecules.
• Generate enzymes with custom active sites.

Topics

• Proteina-Complexa
• Protein Binder Design
• Generative Models
• Flow-Matching
• Enzyme Design

Code references

• NVIDIA-Digital-Bio/Proteina-Complexa

Best for: AI Engineer, Machine Learning Engineer, Research Scientist

Related on AIssential

• From discovery to design: in conversation with Ali Madani (Profluent) — AI-driven language models can design novel, functional proteins from scratch, accelerating biological evolution for therapeutic applications.
• Co-Diffusion: An Affinity-Aware Two-Stage Latent Diffusion Framework for Generalizable Drug-Target Affinity Prediction — Co-Diffusion enhances drug-target affinity prediction in cold-start scenarios using a two-stage, affinity-aware latent diffusion framework.
• Enhancing success rates in therapeutic antibody design through generative models — DualGPT-AB is a dual-stage GPT framework optimizing therapeutic antibody design for specificity and developability.
• AlphaFold hits ‘next level’: the AI database now includes protein pairing — The AlphaFold protein-structure database has been significantly expanded to include predictions of protein complexes, specifically 1.7 million "homodimers," which are crucial for understanding protein function.
• Bridging the phenotype-target gap for molecular generation via multi-objective reinforcement learning — ExMolRL combines phenotype-guided generation with target-aware reinforcement learning for de novo drug discovery.
• Is AI Drug Discovery Becoming a Data Infrastructure Race? — AI drug discovery's primary bottleneck is the availability of diverse, relevant biological training data, not model architecture or compute.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by NVIDIA Technical Blog.