How should I interpret the PLL scores provided by the skill?

Normalized PLL scores above 0.0 generally indicate 'natural-like' sequences. Scores above 0.2 are considered very natural, while scores below -0.5 may indicate unnatural or potentially unstable designs.

Can I use this skill for 3D protein structure prediction?

No, this skill is focused on sequence-level analysis and embeddings. For structure prediction, you should use specialized skills like 'chai' or 'boltz' mentioned in the protein-design-skills repository.

What is the primary purpose of the ESM2 skill?

The ESM2 skill is designed to evaluate protein sequences using large language models, providing metrics like sequence plausibility (PLL scores) and vector representations (embeddings) for downstream analysis.

What hardware do I need to run this skill?

While small models can run on CPU, it is highly recommended to use a GPU with at least 8GB VRAM (24GB A10G recommended) for the 650M or 3B parameter models, either locally or via Modal.

ESM2 Protein Analysis

Name: ESM2 Protein Analysis
Author: adaptyvbio

byadaptyvbio

•

데이터 과학 및 ML

Analyzes protein sequences using the ESM2 language model to generate high-quality embeddings and evaluate biological plausibility.

The ESM2 skill integrates Meta's state-of-the-art protein language model into Claude Code, enabling computational biologists to perform deep sequence analysis. It allows users to compute pseudo-log-likelihood (PLL) scores to rank designs, generate high-dimensional embeddings for clustering, and predict variant effects through zero-shot learning. By streamlining the transition from sequence design to quality control, this skill helps identify natural-like protein candidates and evaluate sequence-function relationships across various model scales, from 8M to 3B parameters.

주요 기능

01Compute Pseudo-Log-Likelihood (PLL) scores for sequence ranking

02Support for multiple ESM2 model sizes (8M, 35M, 650M, and 3B parameters)

03Standardized quality control thresholds for sequence plausibility

0475 GitHub stars

05Perform zero-shot variant effect prediction and mutation scanning

06Generate high-dimensional embeddings for sequence clustering

사용 사례

01Predicting the impact of single-point mutations on protein function

02Analyzing sequence diversity and functional relationships using embeddings

03Filtering synthetic protein designs by biological plausibility scores

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add adaptyvbio/protein-design-skills esm

For use in Claude.ai and ChatGPT

Download Skill