IPSAE

IPSAE is a scoring program for protein-protein interfaces in cofolded complexes from the Dunbrack Lab at the Fox Chase Cancer Center. It takes a predicted complex along with its per-residue confidence scores and the predicted aligned error (PAE) matrix and reports five complementary interface-quality scores for a selected binder-target chain pair. This toolkit runs IPSAE through a single registered tool that returns the ipSAE score together with pDockQ, pDockQ2, the local interaction score (LIS), and interface pTM.

Dunbrack Lab

License: IPSAE is open source and free for academic and commercial use under an MIT license. Please refer to the license for full terms.

Proto is not affiliated with the Dunbrack Lab. This toolkit is open source and builds on the implementation produced by this organization. Product names, logos, and trademarks are the property of their respective owners.

---

GitHub 240 GitHub 240 Preprint Preprint Cite Cite Tool Source Tool Source Open as Notebook Open as Notebook Open on Proto Open on Proto

DunbrackLab/IPSAE

Scoring function for interprotein interactions in AlphaFold2 and AlphaFold3

240 stars

View repo

R=es ipSAE loquuntur: What’s wrong with AlphaFold’s ipTM score and how to fix it

Roland L. Dunbrack

bioRxiv (2025)

Read preprint

@article{dunbrack_2025_ipsae,
  title={R\={e}s ipSAE loquuntur: What's wrong with AlphaFold's ipTM score and how to fix it},
  author={Dunbrack, Roland L.},
  journal={bioRxiv},
  year={2025},
  doi={10.1101/2025.02.10.637595},
  url={https://www.biorxiv.org/content/10.1101/2025.02.10.637595v2},
}

@article{zhu_2023_pdockq2,
  title={Evaluation of AlphaFold-Multimer prediction on multi-chain protein complexes},
  author={Zhu, Wensi and Shenoy, Aditi and Kundrotas, Petras and Elofsson, Arne},
  journal={Bioinformatics},
  volume={39},
  number={7},
  pages={btad424},
  year={2023},
  doi={10.1093/bioinformatics/btad424},
}

@article{bryant_2022_pdockq,
  title={Improved prediction of protein-protein interactions using AlphaFold2},
  author={Bryant, Patrick and Pozzati, Gabriele and Elofsson, Arne},
  journal={Nature Communications},
  volume={13},
  number={1},
  pages={1265},
  year={2022},
  doi={10.1038/s41467-022-28865-w},
}

@article{kim_2024_lis,
  title={Enhanced Protein-Protein Interaction Discovery via AlphaFold-Multimer},
  author={Kim, Ah-Ram and Hu, Yanhui and Comjean, Aram and Rodiger, Jonathan and Mohr, Stephanie E. and Perrimon, Norbert},
  journal={bioRxiv},
  year={2024},
  doi={10.1101/2024.02.19.580970},
  url={https://www.biorxiv.org/content/10.1101/2024.02.19.580970v1},
}

Copy citation

proto-bio/proto-tools/proto_tools/tools/structure_scoring/ipsae

View source

Open Notebook

Open notebook

Coming soon!

Run this tool directly in Proto with no setup required.

Function	Description
run_ipsae_scoring()	Score a cofolded protein complex with IPSAE (Dunbrack 2025), computing ipSAE, pDockQ2, LIS, pDock…	Docs Source

Background

IPSAE (Dunbrack, 2025) is a single program that computes five interface-quality scores for a cofolded protein complex from the structure together with its per-residue pLDDT and pairwise PAE matrix. The primary metric, ipSAE, recomputes an interface pTM-style score from PAE alone while replacing the chain-length-based reference distance with an adaptive reference distance derived from the size of the well-predicted interface. The published benchmark reports that ipSAE separates true and false complexes more efficiently than AlphaFold’s interface pTM, in particular for domain-domain and domain-peptide interactions inside larger constructs that contain disordered or accessory regions.

Alongside ipSAE the program computes four additional interface-quality scores that the literature has converged on for cofolded complex assessment. The pDockQ2 score (Zhu et al., 2023) estimates the quality of each interface in a multimer from interface pLDDT and PAE-derived signal. The pDockQ score (Bryant et al., 2022) is an earlier variant that uses interface pLDDT and the logarithm of the number of contacts. The local interaction score (Kim et al., 2024) reports the fraction of interface residues with low mean cross-chain PAE. The interface pTM (iptm_d0chn) reports the original AlphaFold-style interface pTM recomputed from PAE with the chain-length reference distance.

Learning Resources

• DunbrackLab/IPSAE (Dunbrack Lab, Fox Chase Cancer Center). Official repository and the source of the reference scoring script that this toolkit invokes.

Tools

IPSAE Interface Scoring (ipsae-scoring)

Scores a cofolded protein complex by computing ipSAE, pDockQ2, LIS, pDockQ, and interface pTM for a designated binder chain against one or more target chains. The tool takes a Structure with per-residue pLDDT in the B-factor column and the PAE matrix attached at structure.metrics["pae"], runs the IPSAE scoring program, and returns the headline scores together with a full per-chain-pair breakdown.

API Reference

Source

Input: IPSAEScoringInput

structure · Structure · required

Cofolded complex with per-residue pLDDT in the B-factor column and the PAE matrix attached at structure.metrics['pae'] as a square list[list[float]].

Show Structure

structure · string · required

Raw structure content in PDB or CIF format.

structure_format · string

Format of the content string (auto-detected if omitted).

b_factor_type · BFactorType · default: "unspecified"

What the B-factor column represents.

source · string

Optional source identifier (filepath or tool name).

metrics · Metrics

Associated metrics (e.g., pLDDT, pTM scores, per-chain lists, pairwise matrices). None values are stripped at construction.

binder_chain · SingleChainSelection · required

Single-character chain ID of the binder.

Show SingleChainSelection

chain · string · required

The selected chain ID.

target_chains · ChainSelection · required

Target chain ID(s).

Show ChainSelection

chains · List[string] · required

Chain IDs in the selection.

Source

Config: IPSAEScoringConfig

pae_cutoff · number · default: "10.0"

PAE cutoff in Å for interface residue detection.

distance_cutoff · number · default: "10.0"

CA-CA distance cutoff in Å for contact detection.

verbose · integer · default: "0"

Verbosity level (0=quiet, 1=info, 2=debug, 3=raw subprocess stderr). True is coerced to 1 and False to 0.

device · string · default: "cpu"

Device to run the tool on.

timeout · integer · default: "600"

Maximum execution time in seconds. None waits indefinitely.

seed · integer

Random seed. When set, tools run reproducibly up to small GPU float noise (see BaseToolOutput.approx_equal), and the seed participates in cache keys. When None, cacheable seed-sensitive tools skip cache until seeded.

Source

Output: IPSAEScoringOutput

metrics · IPSAEMetrics · required

Scalar metrics plus per-chain-pair breakdown.

Show IPSAEMetrics

chain_pair_results · List[ChainPairScores]

Full per-chain-pair breakdown.

primary_metric · string

Name of the metric that best summarizes the result overall (e.g. "avg_plddt" for AlphaFold2). Used by downstream UI and reporting to pick a headline value.

Metrics

Metric	Type	Range	Availability
ipsae	float	0.0 to 1.0	always
pdockq2	float	0.0 to 1.5	always
lis	float	0.0 to 1.0	always
pdockq	float	0.0 to 1.0	always
iptm_d0chn	float	0.0 to 1.0	always

Applications

This tool is appropriate for ranking and filtering cofolded complexes from structure-prediction tools such as AlphaFold 3, Chai-1, Boltz, or Protenix. Representative applications include scoring candidate protein binders from a design pipeline, identifying the most promising poses in a multi-chain prediction ensemble, and any analysis that benefits from multiple complementary interface-quality scores in a single call rather than running several scoring programs in sequence.

Usage Tips

• The PAE matrix is required and must be attached at structure.metrics["pae"] as a square list[list[float]]. The dimension should match the total residue count of the structure. The input is rejected when the matrix is missing or not square.
• Per-residue pLDDT must be supplied via the B-factor column. Structure predictors in proto-tools return the correct b_factor_type automatically, and Structure.from_file() auto-detects it for AlphaFold DB and ModelArchive files. For manually provided structures from other sources, pass b_factor_type=BFactorType.PLDDT (raw 0 to 100) or BFactorType.NORMALIZED_PLDDT (0 to 1) explicitly. The input is rejected when b_factor_type is any other value, since pDockQ and pDockQ2 would otherwise be computed incorrectly.
• Missing or duplicated chains are rejected. The binder chain must not also appear in target_chains, and every requested chain must be present in the structure. Single-character chain identifiers are also required because IPSAE reads PDB-format input. Multi-character mmCIF chain labels should be shortened with Structure.to_pdb_with_chain_mapping() before scoring.
• The tool implementation does not check that the PAE matrix is aligned with the structure residue by residue. It only confirms the matrix is square and two-dimensional. The caller is responsible for ensuring the PAE rows and columns match the residue order in the structure. A misaligned PAE matrix produces silently meaningless scores rather than an error.
• The top-level scores report the symmetric maximum value for the binder-target interface. The full directional and symmetric breakdown for every chain pair is available on result.metrics.chain_pair_results. When no symmetric pair matches the binder-target combination, the top-level scores fall back to 0.0 and a warning is logged.
• pae_cutoff and distance_cutoff control the interface definition. Both default to 10.0 Å. Lower values define a tighter interface and reduce the number of residues contributing to each score. Use a tighter cutoff when comparing interfaces of similar overall size or when assessing buried interfaces.
• An empty interface returns zeros for every score. When no residues fall within both cutoffs across the binder-target chain pair, every metric is 0.0. Verify the chain identifiers and cutoff values before interpreting an all-zero result as a poor interface.

Toolkit Notes

These apply to every IPSAE tool in this toolkit (ipsae-scoring).

• Outputs are returned as typed metric objects. Each IPSAEMetrics result carries the five top-level scores, the chain_pair_results breakdown for every chain pair, and the headline primary_metric (ipsae). Results can be exported to JSON through the standard export method.

Example notebook: See the full working example for a copy-paste-ready walkthrough.

Infrastructure Guides

The following guides cover how to run tools efficiently and at scale.

Tool Persistence

Keep a tool’s model warm across calls instead of reloading it every invocation.

Device Management

How GPUs are allocated to tools and how to target specific devices.

Parallel Execution

Fan a batch of inputs out across multiple GPUs.

Cloud Inference

Run tools on managed cloud infrastructure with no local setup.