reactifptm

A reimplementation of the actifpTM score for assessing AlphaFold-Multimer interface confidence.

actifpTM is a variant of ipTM that restricts the per-residue alignment score to interface contacts only — pairs of residues whose Cβ–Cβ distance falls under a contact threshold. This focuses the score on the residues that actually mediate the predicted interaction.

Installation

pip install reactifptm

Or, from a clone:

git clone https://github.com/adamsimpkin/reactifptm
cd reactifptm
pip install -e .

Quickstart

Command line

reactifptm scores.json model.pdb -o results.json

scores.json is a PAE file (any of .json, .npz, .npy, or .pkl containing the predicted aligned error matrix). model.pdb is the corresponding predicted structure (.pdb or .cif).

Output:

reactifpTM: 0.951

pairwise reactifpTM (directional):
  A-B: 0.951
  B-A: 0.940

pairwise reactifpTM (max of both directions):
  A-B: 0.951

Python

from reactifptm import Reactifptm

r = Reactifptm("scores.json", "model.pdb")
overall, pairwise = r.compute_reactifptm()

print(overall)                       # global reactifpTM
print(pairwise)                      # {"A-B": 0.951, "B-A": 0.940}
print(r.reactifptm_pairwise_max)     # {"A-B": 0.951}

r.save_results("results.json")

Output

• reactifptm — the global reactifpTM score, computed over interface contacts across the entire complex.
• reactifptm_pairwise — directional pairwise scores. A-B is the best-aligned residue from chain A's perspective onto its partner; B-A is the same from chain B's side. Useful for diagnosing which side of an interface dominates the score.
• reactifptm_pairwise_max — per unordered pair, the max of the two directional scores. This is what you'd typically compare against reference single-direction implementations.

Options

Development

pip install -e ".[dev]"
pytest

The test suite verifies reactifPTM values against reference actifptm fields shipped in tests/data/.

License

MIT — see LICENSE.