What should I do if my design campaign has a low pass rate?

The skill includes failure recovery trees that suggest specific fixes, such as lowering diffusion noise scales, increasing binder length, or switching to all-atom design tools like BoltzGen.

Can this skill predict actual binding affinity?

No. Research indicates that individual metrics are primarily pre-screening filters. While they eliminate poor designs, composite scoring is required for meaningful ranking, and experimental validation remains necessary.

Does it check for manufacturability?

Yes, it includes checks for expression risks such as odd cysteine counts, deamidation motifs (NG/NS/NT), and hydrophobic runs that could lead to aggregation.

Which software tools does this skill support?

The skill provides thresholds and logic compatible with AlphaFold2, Chai, Boltz, PyRosetta, BioPython, and ESM2-based metrics.

Protein Design Quality Control

Name: Protein Design Quality Control
Author: adaptyvbio

byadaptyvbio

•

Data Science & ML

Evaluates and filters protein designs using research-backed metrics for binding, expression, and structural integrity.

The protein-qc skill provides a comprehensive framework for validating de novo protein designs through multi-stage filtering and composite scoring. It helps researchers move beyond individual metrics—which often have weak predictive power—by implementing standardized thresholds for structural confidence (pLDDT), binding quality (ipTM), and biophysical plausibility. With built-in failure recovery trees and diagnostic commands, it streamlines the identification of high-quality candidates for experimental validation while flagging sequence liabilities like deamidation motifs, unpaired cysteines, and aggregation risks.

Key Features

01Multi-stage filtering pipelines for pLDDT, ipTM, PAE, and scRMSD thresholds.

0275 GitHub stars

03Composite scoring models for advanced design ranking and experimental prioritization.

04Biophysical property analysis including instability index, GRAVY, and pI.

05Sequence liability detection for deamidation, proteolysis, and unpaired cysteines.

06Failure recovery trees for diagnosing and fixing poor design campaign results.

Use Cases

01Automating quality control for binder design campaigns targeting specific hotspots.

02Filtering massive libraries of generated proteins to select the top 1% for lab synthesis.

03Diagnosing failures in protein folding predictions and adjusting diffusion model parameters.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add adaptyvbio/protein-design-skills protein-qc

For use in Claude.ai and ChatGPT

Download Skill