AlphaFold2-Multimer-NIM

Tutorial to run AlphaFold2-Multimer NIM on HiPerGator

AlphaFold2 is a protein structure prediction model from Google DeepMind. AlphaFold2 demonstrates state-of-the-art performance at predicting protein structure form amino acid sequence, besting all other submissions on the Critical Assessment of protein Structure Prediction (CASP).

Features

• Predict protein structure given multiple protein sequences.
• Predict a multiple sequence alignments (MSA) for multiple sequences against a series of protein sequence databases.
• Predict a protein structure given a pre-computed MSA of multiple sequences against protein sequence databases.

For more information about AlphaFold2, see the AlphaFold2 paper in Nature. If you use this NIM or AlphaFold2, make sure to cite the paper:

Jumper, J., Evans, R., Pritzel, A. et al. Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–589 (2021). https://doi.org/10.1038/s41586-021-03819-2

Prerequisites

Minimum Requirements:

• GPU: One NVIDIA GPU with ≥ 32GB VRAM and Compute Capability ≥ 8.0.
• RAM: 64GB.
• CPU: At least 24 cores.
• Storage: 512GB free SSD space for MSA databases.

Recommended for Optimal Performance:

• GPU: One NVIDIA GPU with 80GB VRAM (e.g., A100 80GB).
• RAM: 128GB.
• CPU: At least 36 cores.
• Storage: 512GB fast NVMe SSD space.

Launch AlphaFold2-Multimer NIM on HPG

1. Go to OOD and launch the HiPerGator Desktop.

Note: Remember to update the SLURM Account and QoS to match your group, and adjust the job time accordingly.

2. Start a terminal and run the following commands:

   mkdir -p /blue/groupname/gatorlink/.cache/nim/alphafold2-multimer  # Run only the first time
   export LOCAL_NIM_CACHE=/blue/groupname/gatorlink/.cache/nim/alphafold2-multimer
   mkdir /blue/ufhpc/zhao.qian/.cache/nim/alphafold2-multimer/nvs # Run only the first time
   export MMSEQS_DB_DIR=/blue/ufhpc/zhao.qian/.cache/nim/alphafold2-multimer/nvs
   export MMSEQS_TMP_DIR=/blue/ufhpc/zhao.qian/.cache/nim/alphafold2-multimer/nvs
   ml alphafold2-multimer-nim
   alphafold2-multimer
   start_server

   Note: Since the AlphaFold2 model is quite large (612.47 GB), downloading the model can take up to 2 hours.

   

---

Running Inference

1. Open a New Terminal
   Keep the original terminal running with the launched service.

2. Check Service Status
   In the new terminal, wait until the health check end point returns {"status":"ready"} before proceeding. This may take a couple of minutes. You can use the following command to query the health check.

   curl -X 'GET' \
    'http://localhost:8000/v1/health/ready' \
    -H 'accept: application/json'

   If you would rather check the NIM’s status via python, you can use the requests module.

   import requests
   
   url = "http://localhost:8000/v1/health/ready"  # Replace with the actual URL
   
   headers = {
       "content-type": "application/json"
   }
   try:
       response = requests.get(url, headers=headers)
   
       # Check if the request was successful
       if response.ok:
           print("Request succeeded:", response.json())
       else:
           print("Request failed:", response.status_code, response.text)
   except Exception as E:
       print("Request failed:", E)

3. Navigate to your DESIRED job running directory

   cd /blue/groupname/gatorlink/...

4. Run Inference
   Run inference to get a predicted protein structure for an amino acid sequence using the following command.

   curl -X 'POST' \
    'http://localhost:8000/protein-structure/alphafold2/multimer/predict-structure-from-sequences' \
    -H 'accept: application/json' \
    -H 'Content-Type: application/json' \
    -d '{"sequences": ["MNVIDIAIAMAI", "IAMNVIDIAAI"]}' > output.json

   In python:

   import requests
   import json
   
   url = "http://localhost:8000/protein-structure/alphafold2/multimer/predict-structure-from-sequences"  # Replace with the actual URL.
   sequences = ["MNVIDIAIAMAI", "IAMNVIDIAAI"]  # Replace with the actual sequences you want to perform structure prediction on.
   
   headers = {
       "content-type": "application/json"
   }
   
   data = {
       "sequences": sequences,
       "databases": ["uniref90", "small_bfd"]
   }
   
   response = requests.post(url, headers=headers, data=json.dumps(data))
   
   # Check if the request was successful
   if response.ok:
       print("Request succeeded:", response.json())
   else:
      print("Request failed:", response.status_code, response.text)

5. View the Outputs
   You can use the cat tool print the outputs to the command line as with the following command.

   cat output.json

   For better readability, use jq:

   jq . output.json

   or you can pipe the output directly to jq as in the following command:

   curl -X 'POST' \
    'http://localhost:8000/protein-structure/alphafold2/multimer/predict-structure-from-sequences' \
    -H 'accept: application/json' \
    -H 'Content-Type: application/json' \
    -d '{"sequences": ["MNVIDIAIAMAI", "IAMNVIDIAAI"]}' | jq

---

Endpoints Usage

AlphaFold2-Multimer NIM provides the following endpoints:

• protein-structure/alphafold2/multimer/predict-structure-from-sequences - Predict a protein structure given an input list of amino acide sequences.

• protein-structure/alphafold2/multimer/predict-MSA-from-sequences - Perform a Multiple Sequence Alignment (MSA) and return the MSA and templates for AlphaFold2 inference. This endpoint is useful for batching long-running and CPU-intensive MSA runs prior to structure prediction.

• protein-structure/alphafold2/multimer/predict-structure-from-MSA - Perform structural prediction from an input MSA and templates. This is useful when using a pre-computed or custom/external MSA.

Bash

1. Predict Structure from Multiple Input Sequences (Multimers)

   curl -X 'POST' \
    -i \
    "http://localhost:8000/protein-structure/alphafold2/multimer/predict-structure-from-sequences"  \
    -H 'accept: application/json' \
    -H 'Content-Type: application/json' \
    -d '{"sequences": ["MNVIDIAIAMAI", "IAMNVIDIAAI"], "databases": ["uniref90", "mgnify", "small_bfd"]}'

2. Predict MSA from Multiple Input Sequences (Multimers)

   curl -X 'POST' \
    -i \
    "http://localhost:8000/protein-structure/alphafold2/multimer/predict-msa-from-sequences"  \
    -H 'accept: application/json' \
    -H 'Content-Type: application/json' \
    -d '{"sequences": ["MNVIDIAIAMAI", "IAMNVIDIAAI"], "databases": ["uniref90", "mgnify", "small_bfd"]}'

Python Script

First, you need to load python module:

ml python

1. Predict Structure from Multiple Input Sequences (Multimers), create predict-structure-from-sequences.py.

   python predict-structure-from-sequences.py

2. Predict MSA from Multiple Input Sequences (Multimers), create predict-msa-from-sequences.py.

   python predict-msa-from-sequences.py

3. Predict Protein Structure from MSAs, create predict-structure-from-msa.py.

   python predict-structure-from-msa.py

4. You also can run the python code in Jupyter Notebook

   In the terminal and run the command:

   ml python
   jupyter lab

   After launching JupyterLab, ensure you select the 'python3' kernel before running the notebooks Inference Endpoints.ipynb.

Stopping the NIM Service

To stop the NIM service, simply close the terminal window.

Important Note

Since downloading the AlphaFold2 model can take a very long time and won't affect your next run, you do not need to clean your cache folder. If you would like to save some time, you can copy the model into your cache folder. This may take around 20 minutes instesd of 2 hours.

cp -r /data/ai/tutorial/nim/models/alphafold2-data_v1.1.0 /blue/groupname/gatorlink/.cache/nim/alphafold2-multimer/.

---

Another way to run AlphaFold2-Multimer NIM on HPG

1. Submit a SLURM batch job
   Use sbatch to start the NIM service with GPU resources, and record the name of the node where the service is running.

2. Open a terminal or Jupyter session
   Start an SSH terminal or a Jupyter session using any preferred method (e.g., Open OnDemand, srun, etc.), with minimal resource allocation (no GPU required) to run inference.

3. Run on the same node
   Ensure that the SSH terminal or Jupyter session for inference runs on the same node as the service.

---

References

• AlphaFold2-Multimer Documentation
• AlphaFold2-Multimer on NVIDIA NGC