Skip to main content

ProteinMPNN Execution Guide

The flow

ProteinMPNN is a tool for protein sequence design based on backbone structures. Follow these steps on an E2E GPU Node.

Step 1: Clone Repository​

git clone https://github.com/dauparas/ProteinMPNN.git
cd ProteinMPNN

Step 2: Create Environment​

conda create --name mlfold
source activate mlfold

Step 3: Install Dependencies​

For CPU:

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu

For GPU:

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
conda install cudatoolkit=11.3

Step 4: Load Model​

Update paths and load the pretrained weights:

path_to_model_weights=/mnt/model/vdb/ProteinMPNN/vanilla_model_weights
model_name="v_48_020"
checkpoint_path=$path_to_model_weights/$model_name.pt

Example in Python:

import torch
from protein_mpnn_utils import ProteinMPNN

device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
checkpoint = torch.load(checkpoint_path, map_location=device)
model = ProteinMPNN(num_letters=21, hidden_dim=128, num_encoder_layers=3, num_decoder_layers=3)
model.load_state_dict(checkpoint['model_state_dict'])
model.to(device).eval()
print("Model loaded")

Step 5: Run Design​

Provide a PDB file and specify designed/fixed chains.

Example:

pdb_path = "/mnt/model/vdb/ProteinMPNN/inputs/PDB_monomers/pdbs/5L33.pdb"
designed_chain = "A"
fixed_chain = ""

Adjust design options such as:

num_seqs = 1
sampling_temp = 0.1
batch_size = 1

Run the script to generate designed sequences.

Step 6: Output Results​

The tool outputs designed sequences in FASTA format with scores and sequence recovery rates.

Example output:

>5L33, score=1.6044, designed_chains=['A'], model_name=v_48_020
HMPEEEKAARLFIEALEK...

Examples of PDBs​

You can find Protein Data Bank (PDB) files at the following sources:

Use Cases​

Expanding AlphaFold's Reach​

AlphaFold is revolutionizing structural biology by unlocking previously elusive proteins. This breakthrough has significant applications in multiple fields:

Advancing Medicine​

  • Scientists are leveraging AlphaFold to develop novel treatments for diseases such as cancer, Alzheimer's, and antibiotic resistance.
  • AI-driven protein structure predictions accelerate drug discovery and precision medicine.

Fighting Plastic Pollution​

  • Researchers are engineering enzymes that efficiently break down plastic waste, offering a potential solution to environmental crises.
  • These advancements could lead to a sustainable approach to plastic degradation and waste management.

Improving Global Food Security​

  • Mapping plant proteins helps in developing drought-resistant crops.
  • AI-driven insights contribute to more sustainable agricultural practices, enhancing global food supply.

AI-Driven Life Sciences​

Machine learning is being integrated with physics-based simulations to push the boundaries of biological research:

  • Curing diseases in months, not years – Accelerating the development of life-saving treatments.
  • Beating superbugs before they strike – Predicting antibiotic resistance mechanisms to stay ahead of evolving pathogens.
  • Engineering enzymes that devour plastic waste – Transforming waste management through bioengineering solutions.

AlphaFold's potential in structural biology, drug discovery, and AI-driven life sciences is reshaping how we approach global challenges, making groundbreaking scientific advancements accessible in record time.