rst-ipsae

name	rst-ipsae
description	Score predicted protein-protein interfaces with rust_simulation_tools — compute ipSAE, pDockQ, pDockQ2, LIS, and ipTM from a PDB/CIF structure plus pLDDT and PAE, or directly from in-memory arrays. Use for "ipSAE", "pDockQ", "interface confidence", "scoring AlphaFold/Boltz/Chai complexes", "PAE-based interface score", or "LIS".

ipSAE interface scoring

Computes interface-quality metrics for predicted complexes (AlphaFold-Multimer, Boltz, Chai, etc.) from per-residue confidence (pLDDT) and the predicted aligned error (PAE) matrix. Metrics: pDockQ, pDockQ2, LIS, ipTM, and ipSAE.

From a structure file

import numpy as np
from rust_simulation_tools import compute_ipsae

# pLDDT: per-residue confidence on a 0-100 scale, shape (N,)
plddt = np.load("confidences.npz")["plddt"] * 100.0   # multiply if yours is 0-1 (e.g. Boltz)

# PAE: predicted aligned error (N, N), flattened row-major to 1-D
pae = np.load("pae.npz")["pae"].flatten()             # shape (N*N,)

results = compute_ipsae("model.pdb", plddt, pae)      # PDB or CIF

results is a dict with two keys:

• "directed_pairs" — every ordered chain pair (A→B and B→A separately)
• "max_pairs" — one entry per unordered pair, keeping the higher-ipSAE direction

for p in results["directed_pairs"]:
    print(p["chain1"], "->", p["chain2"],
          p["pDockQ"], p["pDockQ2"], p["LIS"], p["ipTM"], p["ipSAE"])

From in-memory arrays (no file I/O)

Use when coordinates/chains already come from MDAnalysis, BioPython, etc.:

from rust_simulation_tools import compute_ipsae_from_arrays

coords = ...                       # (N, 3) CA / C1' coordinates, Angstroms
chains = ["A", "A", "A", "B", "B"] # per-residue chain label, length N
chain_types = {"A": "protein", "B": "protein"}   # or "nucleic_acid"
results = compute_ipsae_from_arrays(
    coords, chains, chain_types, plddt, pae,
    pdockq_cutoff=8.0,   # contact distance cutoff (default 8 A)
    pae_cutoff=12.0,     # PAE cutoff for LIS/ipSAE (default 12 A)
)

compute_ipsae takes the same optional pdockq_cutoff / pae_cutoff arguments.

Key points

• pLDDT must be on the 0-100 scale — convert 0-1 confidences first.
• PAE must be flattened to 1-D, row-major, length N*N.
• All per-residue arrays share the same length/ordering as the structure.

Full example: examples/example_ipsae.py.