Does this skill require a specific GPU?

Yes, structure prediction is hardware-intensive. A minimum of 24GB VRAM is required, though 40GB (A100) is recommended for complex structures.

Can I use this for high-throughput protein binder validation?

Yes, it is specifically optimized for binder validation, allowing users to check ipTM scores to ensure interface confidence across large design campaigns.

What metrics are used to measure prediction confidence?

The skill provides pLDDT for local confidence, pTM for global structure confidence, and ipTM for the confidence of protein-protein interfaces.

How does Chai-1 compare to AlphaFold2?

Chai-1 is generally faster, does not require a Multiple Sequence Alignment (MSA) for input, and natively supports small molecules (ligands) and nucleic acids (DNA/RNA).

What file formats are supported for input and output?

Input is handled via standard FASTA files (including SMILES for ligands). Output is provided in CIF format, along with JSON and NPY files for confidence scores.

Chai-1 Structure Prediction

Name: Chai-1 Structure Prediction
Author: adaptyvbio

byadaptyvbio

•

데이터 과학 및 ML

Predicts multi-modal molecular structures including protein complexes, ligands, and nucleic acids using the Chai-1 foundation model.

The Chai skill enables Claude to perform high-accuracy molecular structure predictions for proteins, ligands, DNA, and RNA without requiring Multiple Sequence Alignment (MSA). It streamlines the validation of designed protein binders, predicts protein-ligand interactions via SMILES strings, and provides essential confidence metrics like pLDDT and ipTM. This skill is particularly useful for researchers needing a fast, versatile alternative to AlphaFold2 that supports diverse molecular entities and high-throughput workflows through Modal or the Chai API.

주요 기능

01High-throughput inference via Chai API or Modal-powered GPU execution

02Automated calculation of confidence metrics including pTM, ipTM, and pLDDT

03Multi-modal prediction for proteins, small molecules (SMILES), and DNA/RNA

04Comprehensive binder validation and protein-protein interface scoring

0575 GitHub stars

06MSA-free structural analysis for significantly faster results than AlphaFold2

사용 사례

01Modeling large-scale protein complexes and nucleic acid interactions for structural biology

02Predicting the 3D configuration of protein-ligand complexes for drug discovery

03Validating the binding affinity and interface quality of de novo designed protein binders

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add adaptyvbio/protein-design-skills chai

For use in Claude.ai and ChatGPT

Download Skill