Skip to content
Skills Library
Science & research · Analysis

ipsae

/ipsae

Binder design ranking using ipSAE (interprotein Score from Aligned Errors). Use this skill when: (1) Ranking binder designs for experimental testing, (2) Filtering BindCraft or RFdiffusion outputs, (3) Comparing AF2/AF3/Boltz predictions, (4) Predicting binding success rates, (5) Need better ranking than ipTM or iPAE. For structure prediction, use chai or alphafold. For QC thresholds, use protein-qc.

Licence
MIT
Last repository change
11.06.2026

Assessment and security

100/ 100Class: good
Format25 / 25
Frontmatter with name and description, valid name, description up to 1,024 characters, name matching the folder.
Discoverability25 / 25
Description explains when to use the skill, has a reasonable length and uses third person.
Content35 / 35
At least 150 words and at most 500 lines, headings, step lists, examples or code.
Hygiene15 / 15
No placeholders such as TODO, no Windows paths, with a licence.
Riskno signals detected in the scan
Files in this package
1
Scripts
0
HTML comments
0
URLs in SKILL.md
2
Deprecation

No signals detected. Repository changed on 11.06.2026, without references to retired models or APIs.

Licence

MIT, open: use and modify while retaining author attribution.

What it does and when to use it

Binder design ranking using ipSAE (interprotein Score from Aligned Errors). Use this skill when: (1) Ranking binder designs for experimental testing, (2) Filtering BindCraft or RFdiffusion outputs, (3) Comparing AF2/AF3/Boltz predictions, (4) Predicting binding success rates, (5) Need better ranking than ipTM or iPAE. For structure prediction, use chai or alphafold. For QC thresholds, use protein-qc.

Original author description

Binder design ranking using ipSAE (interprotein Score from Aligned Errors). Use this skill when: (1) Ranking binder designs for experimental testing, (2) Filtering BindCraft or RFdiffusion outputs, (3) Comparing AF2/AF3/Boltz predictions, (4) Predicting binding success rates, (5) Need better ranking than ipTM or iPAE. For structure prediction, use chai or alphafold. For QC thresholds, use protein-qc.

How it works

  1. Prerequisites
  2. Overview
  3. How to run
  4. Key parameters
  5. Output format
  6. Sample output
  7. Decision tree
  8. Recommended thresholds
  9. Batch processing
  10. Verify
Original SKILL.md headings · 175 lines, 618 words
SKILL.md preview
---
name: ipsae
description: >
  Binder design ranking using ipSAE (interprotein Score from Aligned Errors).
  Use this skill when: (1) Ranking binder designs for experimental testing,
  (2) Filtering BindCraft or RFdiffusion outputs,
  (3) Comparing AF2/AF3/Boltz predictions,
  (4) Predicting binding success rates,
  (5) Need better ranking than ipTM or iPAE.

  For structure prediction, use chai or alphafold.
  For QC thresholds, use protein-qc.
license: MIT
category: evaluation
tags: [ranking, scoring, binding]
---

# ipSAE Binder Ranking

## Prerequisites

| Requirement | Minimum | Recommended |
|-------------|---------|-------------|
| Python | 3.8+ | 3.10 |
| NumPy | 1.20+ | Latest |
| RAM | 8GB | 16GB |

## Overview

ipSAE (interprotein Score from Aligned Errors) is a scoring function for ranking protein-protein interactions predicted by AlphaFold2, AlphaFold3, and Boltz1. It separates true from false predicted complexes more reliably than ipTM, which dilutes interface confidence across disordered or accessory regions. In a separate binder meta-analysis (Overath et al. 2025), AF3 ipSAE_min gave a 1.4-fold gain in average precision over the ipAE score that RFdiffusion pipelines commonly filter on.

**Paper**: Dunbrack, "Rēs ipSAE loquuntur: What's wrong with AlphaFold's ipTM score and how to fix it", [bioRxiv 2025.02.10.637595](https://www.biorxiv.org/content/10.1101/2025.02.10.637595v2)

## How to run

### Installation
```bash
git clone https://github.com/DunbrackLab/IPSAE.git
cd IPSAE
pip install numpy
```

### AlphaFold2
```bash
python ipsae.py scores_rank_001.json unrelaxed_rank_001.pdb 15 15
```

### AlphaFold3
```bash
python ipsae.py fold_model_full_data_0.json fold_model_0.cif 10 10
```

### Boltz1
```bash
python ipsae.py pae_model_0.npz model_0.cif 10 10
```

## Key parameters

| Parameter | Description | Recommended |
|-----------|-------------|-------------|
| PAE file | JSON (AF2/AF3) or NPZ (Boltz) | Match predictor |
| Structure file | PDB or CIF structure | Match PAE |
| PAE cutoff | Threshold for contacts | 10-15 |
| Distance cutoff | Max CA-CA distance (A) | 10-15 |

## Output format

Two output files are generated:

**Chain-pair scores** (`_chains.csv`):
```
chain_A,chain_B,ipSAE_min,pDockQ,pDockQ2,LIS,n_contacts,interface_dist
A,B,0.72,0.65,0.58,0.45,42,8.5
```

**Residue-level scores** (`_residues.csv`):
```
chain,resnum,pSAE,pLDDT
A,45,0.85,92.3
A,67,0.78,88.1
```

## Sample output

### Successful run
```
$ python ipsae.py scores_rank_001.json design_0.pdb 10 10
Processing design_0...
Found 2 chains: A, B
Computing ipSAE scores...

Results written to:
  design_0_chains.csv
  design_0_residues.csv

Summary:
  ipSAE_min: 0.72
  pDockQ: 0.65
  LIS: 0.45
  Interface contacts: 42
```

**What good output looks like:**
- ipSAE_min > 0.61 (primary filter)
- pDockQ > 0.5 (supporting metric)
- Reasonable number of interface contacts (20-100)

## Decision tree

```
Should I use ipSAE?
│
├─ What are you ranking?
│  ├─ Designed binders → ipSAE ✓
│  ├─ Natural complexes → ipTM is fine
│  └─ Single proteins → Not applicable
│
├─ What predictor did you use?
│  ├─ AlphaFold2 → ipSAE ✓
│  ├─ AlphaFold3 → ipSAE ✓
│  ├─ Boltz1 → ipSAE ✓
│  ├─ Chai → ipSAE (use PAE output)
│  └─ ESMFold → Not applicable (no PAE)
│
└─ Why ipSAE over ipTM?
   ├─ Different length constructs → ipSAE ✓
   ├─ Designs with disordered regions → ipSAE ✓
   └─ Standard complexes → Either works
```

## Recommended thresholds

| Metric | Standard | Stringent | Use Case |
|--------|----------|-----------|----------|
| ipSAE_min | > 0.61 | > 0.70 | Primary filter |
| LIS | > 0.35 | > 0.45 | Interface quality |
| pDockQ | > 0.5 | > 0.6 | Supporting |

## Batch processing

```python
import subprocess
import os
from pathlib import Path

def score_designs(pae_dir, struct_dir, output_dir):
    """Score all designs in a directory."""
    Path(output_dir).mkdir(exist_ok=True)

    for pae_file in Path(pae_dir).glob("*_scores*.json"):
        name = pae_file.stem.replace("_scores_rank_001", "")
        struct_file = Path(struct_dir) / f"{name}.pdb"

        if struct_file.exists():
            subprocess.run([
                "python", "ipsae.py",
                str(pae_file),
                str(struct_file),
                "10", "10"
            ])
```

---

## Verify

```bash
ls *_chains.csv | wc -l  # Should match number of predictions
```

---

## Troubleshooting

**Low scores for good designs**: Check PAE/distance cutoffs
**Missing output**: Verify PAE file format matches predictor
**Inconsistent scores**: Use same cutoffs across all designs

### Error interpretation

| Error | Cause | Fix |
|-------|-------|-----|
| `KeyError: 'pae'` | Wrong PAE format | Check if AF2/AF3/Boltz format |
| `FileNotFoundError` | Structure not found | Verify file paths |
| `ValueError: no contacts` | No interface detected | Check chain IDs, reduce cutoffs |

---

**Next**: Select top designs (ipSAE_min > 0.61) → experimental validation.

Versions

RepositoryScoreLicence
This page · adaptyvbio/protein-design-skills100MIT
FreedomIntelligence/OpenClaw-Medical-Skills100MIT

Database snapshot: 13.11.2026 · package version fac3a87be8e7 (SHA-256 digest of the entire package)

Downloads in the last 30 days: 1 · ranking

Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.

Similar skills

  • 100binder-design
    Guidance for choosing the right protein binder design tool. Use this skill when: (1) Deciding between BoltzGen, BindCraft, or RFdiffusion, (2) Planning a binder design campaign, (3) Understanding trade-offs between different approaches, (4) Selecting tools for specific target…
  • 100bindcraft
    End-to-end binder design using BindCraft hallucination. Use this skill when: (1) Designing protein binders with built-in AF2 validation, (2) Running production-quality binder campaigns, (3) Using different design protocols (fast, default, slow), (4) Need joint backbone and…
  • 95protenix
    Structure prediction with Protenix, an open AlphaFold3 reproduction. Use this skill when: (1) Predicting complex structures with an AF3-class model, (2) Wanting an open alternative to AF3 alongside Boltz and Chai, (3) Validating designed binder-target complexes. For QC…
  • 100bindcraft
    End-to-end binder design using BindCraft hallucination. Use this skill when: (1) Designing protein binders with built-in AF2 validation, (2) Running production-quality binder campaigns, (3) Using different design protocols (fast, default, slow), (4) Need joint backbone and…
  • 100boltz
    Structure prediction using the locally registered Boltz-2 installation. Use this skill when: (1) Predicting protein complex structures, (2) Validating designed binders, (3) Need open-source alternative to AF2, (4) Predicting protein-ligand complexes, (5) Using local GPU…
  • 100protein-binder-design
    Design new proteins that bind a chosen surface, using BindCraft's AlphaFold2-guided hallucination or the RFdiffusion backbone plus ProteinMPNN sequence pipeline. Use this skill to specify a target epitope by hotspot residue, trim a receptor to the region worth designing against,…

Feedback and reports

Your message goes to Bartosz Sokoliński. An email address is needed only for a reply. See the privacy policy.