Pre-print available on bioRxiv: AlphaBind, a Domain-Specific Model to Predict and Optimize Antibody-Antigen Binding Affinity
Table of contents
To build the image, you will need a version of Docker with support for buildkit Dockerfile syntax >= v1.10.0. Installation of Docker on various platforms is beyond the scope of this documentation, but instructions can be found in the official Docker documentation.
In order to download the model weights for the ESM-2nv model from NGC, you will need a free NGC account and API Key. If you do not already have these, create them by following NVIDIA's NGC Account and API Key Configuration documentation.
Securely set the NGC_CLI_API_KEY and NGC_CLI_ORG environment variables on the host system using the credentials obtained above.
If performing this step manually, security best practice is to avoid persisting these values in your shell history. One approach to doing so is to populate the provided ngc_secrets.env.template file with your API credentials, rename it to ngc_secrets.env, then export the environment variables in that file by running:
set -a
source ./ngc_secrets.env
set +a
Important
If you previously set the environment variables in the preceding section using the source ./ngc_secrets.env method, the following command must be run in that same shell session (or a subshell thereof).
# Extract the alphabind Python package version from our pyproject.toml
ALPHABIND_VERSION=$(sed -n 's/.*version = "\([^"]*\)".*/\1/p' ./alphabind/pyproject.toml)
docker build --secret id=NGC_CLI_API_KEY --secret id=NGC_CLI_ORG -t alphabind:latest -t alphabind:${ALPHABIND_VERSION} .
We recommend that most users start with our two tutorial notebooks to familiarize themselves with terminology and usage details.
This tutorial details a usage example for fine-tuning the AlphaBind pre-trained checkpoint using a third-party dataset. Additional details are available in the tutorial's accompanying README.
Important
This tutorial depends on prior successful completion of Tutorial 1.
This tutorial details a usage example for optimizing a parental sequence against a target using the fine-tuned model trained in Tutorial 1. Additional details are available in the tutorial's accompanying README.
Pre-print available on bioRxiv: AlphaBind, a Domain-Specific Model to Predict and Optimize Antibody-Antigen Binding Affinity
Figure plotting code, data, and representative production commands associated with our manuscript can be found in the manuscript directory. Note that a few legend naming conventions were cosmetically aliased for our pre-print, downstream of the plotting code in this repository, but figure content is identical.
See LICENSE.
All product names, logos, and brands mentioned in this documentation are property of their respective owners. "NVIDIA", "NGC", and "BioNeMo" are trademarks or registered trademarks of NVIDIA Corporation. "Docker" is a trademark or registered trademark of Docker, Inc. The use of these names, logos, and brands does not imply endorsement.
© A-Alpha Bio, Inc. 2024. All rights reserved.