Use Case Description
In drug discovery, even if you know which protein to target that would treat a disease, designing a therapeutic molecule that specifically binds that protein is 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. Here, we bring generative AI to bear on the problem, pre-optimizing molecules and screening 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.
Experience Walkthrough
-
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 AlphaFold2, which predicts the initial 3D structure of the target protein. AlphaFold2 also requires a multi-sequence alignment, which can be generated with an accelerated MSA NIM.
-
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.
-
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.
-
Finally, AlphaFold2-Multimer 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.
-
The Generative Protein Binder Design NVIDIA BioNeMo 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 AlphaFold2, which predicts the initial 3D structure of the target protein. AlphaFold2 also requires a multi-sequence alignment, which can be generated with an accelerated MSA NIM.
-
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.
-
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.
-
Finally, AlphaFold2-Multimer 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.
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.
Minimum System Requirements
Hardware
- GPU: 4 or more NVIDIA L40s, A100, or H100 GPUs, each with at least 48 GB of VRAM
- CPU: x86_64 architecture only for this release, with 24 physical CPU cores
- Storage: 1300GB of fast NVMe SSD storage
- System Memory: 64GB
Software
- Operating System: Ubuntu 20.04 or newer
- NVIDIA Driver version: 535 or newer
- NVIDIA CUDA® version: 12.4 or newer
- NVIDIA Container Toolkit version: 1.15.0 or newer
- Docker version: Docker version 26 or newer
- Python Version 3.11+
Example Walkthrough with Sample Input/Output
See a complete example of how to get started with this blueprint on the NVIDIA BioNeMo Blueprints GitHub repository
Technical Considerations
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. While AI is making great strides in predicting protein sequences and structures that should bind to target proteins, it’s not perfect.
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. When downloaded or used in accordance with our terms of service, developers should work with their supporting model team to ensure the models meet requirements for the relevant industry and use case and addresses unforeseen product misuse. For more detailed information on ethical considerations for the models, please see the Model Card++ Explainability, Bias, Safety & Security, and Privacy Subcards. Please report security vulnerabilities or NVIDIA AI Concerns here.
License
Use of the models in this Blueprint is governed by the NVIDIA AI Foundation Models Community License.
