ipSAE
Score predicted binder interfaces with ipSAE.
Results
Loading starting example…
Method & evidence
Calculate native ipSAE v4 directional interface scores over a bounded cohort of checksum-bound predicted complexes and per-token PAE evidence.
Scope & limitations
- Use checksum-matched PAE from the same prediction that produced the complex.
Interpreting results
- Higher ipSAE can support prioritization within a consistently predicted target cohort.
- Calibrate scores within target and prediction method; avoid assuming one global threshold.
- ipSAE is confidence-derived evidence, not an affinity measurement.
References
Input requirements & outputs
Input checks
- PDB or mmCIF predicted complex
- JSON or NPZ per-token PAE with matching checksum
- Up to 25 rows in one native batch
Expected results
- Directional interface score table — Compare every chain-pair ipSAE metric without collapsing directionality.
- Native score evidence — Retain tool version, cutoffs, source checksums, row failures, and result digest.
