What is the primary advantage of ipSAE over ipTM?

ipSAE offers roughly 1.4x higher precision in identifying true binders compared to ipTM, especially for designed binders and constructs with disordered regions.

Which protein structure predictors are supported?

This skill supports AlphaFold2 (.json), AlphaFold3 (.json/.cif), Boltz1 (.npz/.cif), and Chai-1 outputs.

Can I use this skill for single protein analysis?

No, ipSAE is specifically designed for ranking interprotein interactions and binder-target complexes rather than single protein chains.

What is considered a good ipSAE score?

For primary filtering, a score of ipSAE_min > 0.61 is considered the standard threshold, while > 0.70 is used for stringent selection.

Does the skill provide residue-level detail?

Yes, it generates both chain-pair summary files and residue-level CSVs containing pSAE and pLDDT scores.

ipSAE Binder Ranking

Name: ipSAE Binder Ranking
Author: adaptyvbio

byadaptyvbio

•

Data Science & ML

Ranks and evaluates protein binder designs using the ipSAE scoring function to identify high-confidence protein-protein interactions.

The ipSAE skill provides specialized capabilities for ranking protein-protein interaction (PPI) predictions generated by AlphaFold2, AlphaFold3, and Boltz1. By utilizing the interprotein Score from Aligned Errors (ipSAE), it offers significantly higher precision in identifying successful binders compared to traditional ipTM or iPAE metrics. This skill is ideal for protein engineers needing to filter RFdiffusion or BindCraft outputs, providing detailed chain-pair and residue-level metrics to prioritize candidates for experimental validation.

Key Features

01Support for AlphaFold2, AlphaFold3, and Boltz1 model outputs

02Automated batch processing for large design libraries

03Detailed interface metrics including pDockQ, LIS, and contact counts

04High-precision binder ranking with 1.4x better performance than ipTM

0575 GitHub stars

06Residue-level scoring for precise interface quality assessment

Use Cases

01Comparing prediction quality across different protein structure models

02Ranking de novo protein binder designs for experimental testing

03Filtering structural outputs from RFdiffusion and BindCraft

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add adaptyvbio/protein-design-skills ipsae

For use in Claude.ai and ChatGPT

Key Features

01Support for AlphaFold2, AlphaFold3, and Boltz1 model outputs

02Automated batch processing for large design libraries

03Detailed interface metrics including pDockQ, LIS, and contact counts

04High-precision binder ranking with 1.4x better performance than ipTM

0575 GitHub stars

06Residue-level scoring for precise interface quality assessment

Use Cases

01Comparing prediction quality across different protein structure models

02Ranking de novo protein binder designs for experimental testing

03Filtering structural outputs from RFdiffusion and BindCraft

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add adaptyvbio/protein-design-skills ipsae

For use in Claude.ai and ChatGPT