Description:

In drug discovery, even after a target protein has been identified for treating a disease, designing a therapeutic molecule that specifically binds that protein remains a staggering challenge. Imagine searching for a single, perfectly shaped key in a warehouse of nearly infinite keys—each with a unique three-dimensional shape. This isn't just a metaphor; for a protein of length ‘n’, there can be 20^n possible sequences, each capable of adopting countless conformations. Since the average human protein is 430 amino acids, this represents 20^430 possible sequences, a practically infinite number and more than the number of atoms in the universe (10^80). This potential diversity is so important in the evolution of life but presents a challenge for researchers. In traditional workflows, this complexity means painstaking trial and error—iterating through thousands of candidates, each synthesis and validation round taking months, if not years. The process is expensive, slow, and fraught with uncertainty. Researchers often use educated guesses and hope that a binder emerges from the colossal search. This blueprint applies generative AI to pre-optimize molecules and screen their interaction with the target protein. This BioNeMo blueprint shows how protein binder design can be recast using NIM microservices for protein folding, structure generation, and sequence generation to speed up the development cycle and produce better binders faster. This blueprint is for research and development only.

License/Terms of Use:

GOVERNING HOSTING TERMS: Use of this trial service is governed by the NVIDIA API Trial Terms of Service. Use of this blueprint software is governed by the Apache 2.0, and enables use of separate open source and proprietary software, models, data and services governed by their respective licenses below:

1. Use of the RfDiffusion NIM is governed by the NVIDIA Software License Agreement and Product-Specific Terms for AI Products. Use of the model is governed by the NVIDIA AI Foundation Models Community License Agreement. Additional Information: BSD license.
   
2. Use of the ProteinMPNN NIM is governed by the NVIDIA Software License Agreement and Product-Specific Terms for NVIDIA AI Products. Use of the model is governed by the NVIDIA Community Model License. ADDITIONAL INFORMATION: MIT License.
   
3. Use of the OpenFold3 NIM is governed by the NVIDIA Software License Agreement and Product-Specific Terms for NVIDIA AI Products. Use of the model is governed by the NVIDIA Open Model License. ADDITIONAL INFORMATION: Apache 2.0.
   

Third-Party Community Consideration:

The software components in this blueprint are not owned or developed by NVIDIA. They have been developed and built to a third-party's requirements; see links to Non-NVIDIA Model Cards: OpenFold3, RFdiffusion, and ProteinMPNN.

Deployment Geography:

Global

Architecture Diagram:

What's Included in the Blueprint:

NVIDIA AI Blueprints are customizable AI workflow examples that equip enterprise developers with NIM microservices, reference code, documentation, and a Helm chart for deployment.

Use Case:

Computational biologists, protein engineers, and drug discovery researchers use this blueprint to design novel protein binders against a target protein of interest. It supports teams exploring generative AI–driven alternatives to slow, expensive trial-and-error binder design workflows in therapeutic protein development and other protein engineering applications.

Experience Walkthrough:

1. The Protein Binder Design NVIDIA NIM Agent Blueprint leverages AI models packaged within NIM microservices to design optimized protein sequences and structures. The workflow begins with the user providing an amino acid sequence to OpenFold3, which predicts the initial 3D structure of the target protein. OpenFold3 also requires a multi-sequence alignment, which can be generated with an accelerated MSA NIM.
   
2. The structure of the protein target is then used by RFdiffusion to design a protein binder. At this stage, RFdiffusion generates only the backbone of the protein binder. The model can be steered by the user to explore specific binding interfaces, or hot spot regions of the target protein and identify the most favorable binding configurations according to the user’s desired design constraints.
   
3. Next, ProteinMPNN generates and optimizes amino acid sequences that fit into the RFdiffusion-generated protein backbone, ensuring they exhibit the necessary biochemical properties for effective binding.
   
4. Finally, OpenFold3 is used to validate the interactions and stability of the resulting protein complexes. This integrated approach enables the precise and efficient design of protein binders, facilitating advancements in therapeutic protein development and other protein engineering applications.
   

See a complete example of how to get started with this blueprint on the NVIDIA BioNeMo Blueprints GitHub repository.

RFdiffusion is capable of generating protein backbones, and ProteinMPNN can label the amino acid sequence. The combination yields sequences that should fold into the protein structure that binds to the specified static target protein structure. Note that proteins are flexible and adopt multiple conformations. This is especially true of antibodies where the binding interface is disordered. This poses a challenge for these models.

Release Date:

build.nvidia.com 07/06/2026 via https://build.nvidia.com/nvidia/protein-binder-design-for-drug-discovery
Github 07/06/2026 via https://github.com/NVIDIA-BioNeMo-blueprints/generative-protein-binder-design

Blueprint Version:

2.0.0

Minimum System Requirements:

Hardware Requirements

• At least 100 GB of fast NVMe SSD space
  
• A modern CPU with at least 24 CPU cores
  
• At least 64 GB of RAM
  
• Three or more NVIDIA L40s, A100, or H100 GPUs
  

OS Requirements

• Operating System: Ubuntu 20.04 or newer
  
• NVIDIA Driver version: 580 or newer
  
• NVIDIA CUDA® version: 13.0 or newer
  
• NVIDIA Container Toolkit version: 1.15.0 or newer
  
• Docker version: Docker version 26 or newer
  
• Python Version 3.11+
  

Software Used in This Blueprint:

OpenFold3 NIM, RFdiffusion NIM, and ProteinMPNN NIM.

Ethical Considerations:

NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. Developers should work with their internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse. Please report blueoprint quality, risk, security vulnerabilities or NVIDIA AI Concerns here.