Run Biohub protein-language-model tools for embeddings, scoring, and design workflows.
npx skills add https://github.com/LiorZ/protein-design-skills --skill esm-biohub
Please help me install this Agent Skill. Skill: esm-biohub Source: https://github.com/LiorZ/protein-design-skills/tree/main/skills/esm-biohub Command: npx skills add https://github.com/LiorZ/protein-design-skills --skill esm-biohub
This skill enables researchers to run Biohub's world-model protein suite (ESMC, ESMFold2, ESM3, and SAEs) locally via an Apptainer / Singularity image or through the Biohub Platform, unlocking embeddings, structure predictions, mutation scoring, and de novo design workflows without manual tool integration.
Start by building or obtaining the Esm SIF image, launch the container, and follow the cookbook tutorials to run embeddings, folding, and SAE feature extraction.
High-intent search queries and answers about installing and using this skill.
Run ESM protein models locally for multi-chain structure prediction by building the Apptainer/Singularity image to access ESMFold2 and ESM3 pipelines, enabling folding and design workflows without manual tool integration.
The Biohub ESM suite is used for protein design to generate per-residue embeddings, perform zero-shot mutation scoring, inverse folding, and de novo binder validation across single-chain and multi-chain scenarios.
Yes, you need a Biohub API key to access platform-enabled models when running cloud-based large-scale batch processing, unless you are operating the ESMC, ESMFold2, ESM3, and SAE pipelines entirely locally via the container image.
Yes, ESM3 supports zero-shot mutation scoring with optional MSAs, backbone mutations, and CCD-based ligand interactions, operating across multi-chain scenarios with optional tensor outputs and SAE feature extraction.
Scale protein embedding workflows by using the Biohub Platform API for cloud-hosted model access and batch processing, bypassing local resource limits while maintaining access to ESMC and ESM3 embedding pipelines.
Running ESM models locally requires the Hugging Face cache to store model weights and configurations for the ESMC, ESMFold2, and ESM3 pipelines, ensuring the Apptainer/Singularity image can load protein-language-model tools efficiently.
LiorZDenoise and rebuild atomic coordinates in protein-ligand complexes to generate stochastic ensembles.
adaptyvbioRank binder designs using ipSAE scores from AlphaFold2, AlphaFold3, and Boltz1 predictions.
NVIDIA-BioNeMoGenerate multiple sequence alignments and structural templates for protein sequences.
dralkhAutomate protein design and analysis workflows using the esm Python SDK.
