Metrics and Filters
Choose filter thresholds and interpret the metrics and sequence scores Phi reports.
Metric ranges and thresholds
| Metric | Better | Range | Recommended threshold | passes_all | Source |
|---|---|---|---|---|---|
| pLDDTMean per-residue confidence of the predicted complex. | higher | 0 to 100 | ≥ 80 | ≥ 80 | AlphaFold2 runner, plddt_complex |
| pTMPredicted TM-score of the whole complex. | higher | 0 to 1 | ≥ 0.55 | AlphaFold2 runner, ptm | |
| ipTMPredicted TM-score of the interface alone. | higher | 0 to 1 | ≥ 0.5 | AlphaFold2 runner, iptm | |
| ipSAEInterface confidence from aligned-error blocks; less sensitive to chain length than ipTM. | higher | 0 to 1 | ≥ 0.35 | AlphaFold2 runner, ipsae | |
| iPAEMean predicted aligned error across the interface. | lower | 0 to 31 Å | ≤ 10.85 Å | ≤ 10 Å | AlphaFold2 runner, ipae |
| Binder RMSDCα RMSD between the designed binder and its refolded prediction. | lower | 0 to 20 Å | ≤ 3.5 Å | ≤ 3.5 Å | AlphaFold2 runner, rmsd |
| Binder pLDDTMean per-residue confidence of the binder chain alone. | higher | 0 to 100 | AlphaFold2 runner, plddt_binder | ||
| Ranking scoreThe model's own ranking of its samples, mixing interface and chain confidence. | higher | 0 to 1 | Chai-1, Protenix, OpenDDE runners, ranking_score; OpenFold 3, sample_ranking_score | ||
| gPDEGlobal predicted distance error, in the model's own units; lower is more confident. | lower | 0 to 5 | OpenFold 3, Protenix, OpenDDE runners, gpde | ||
| ConfidenceThe model's overall confidence. | higher | 0 to 1 | Boltz runners, confidence_score | ||
| iPAEMinimum predicted aligned error across the interface, as the refolding oracle reports it. | lower | 0 to 31 Å | RFD4-Proteina and BoltzGen runners, ipae | ||
| scRMSDBackbone RMSD between the design and its refolded prediction. | lower | 0 to 20 Å | RFD4-Proteina runner, refolded_backbone_rmsd; BoltzGen, rmsd | ||
| Hotspot obedienceFraction of the requested hotspot atoms with a binder atom within 4.5 Å. | higher | 0 to 1 | RFD4-Proteina runner, hotspot_obedience | ||
| ClashesFraction of atoms in a steric clash at 1.5 Å. | lower | 0 to 0.2 | RFD4-Proteina runner, clash_metrics | ||
| Sequence adherenceFraction of sequence-conditioned residues the design kept. | higher | 0 to 1 | RFD4-Proteina runner, sequence_adherence | ||
| Motif RMSDAll-atom RMSD of the conditioned region against what was asked for. | lower | 0 to 10 Å | RFD4-Proteina runner, design_motif_allatom_rmsd | ||
| CDR contactsFraction of the antibody's contacting residues that sit in a CDR. | higher | 0 to 1 | RFD4-Proteina runner, antibody_contacts | ||
| Cα RMSD to inputHow far a refined design moved from the structure it started from. | lower | 0 to 20 Å | RFD4-Proteina runner, ca_rmsd_to_input | ||
| HelixFraction of residues in helix. | higher | 0 to 1 | RFD4-Proteina runner, sse_distribution | ||
| SheetFraction of residues in strand. | higher | 0 to 1 | RFD4-Proteina runner, sse_distribution |
Recommended threshold: Dyno's current set, 29 Sept 2026.
passes_all: the API's fixed flag on every persisted design. All three thresholds are required.
Filter sets and sources
Each row uses the source's values in Phi's units. Evidence describes how the source obtained them. Review uses the selected set to mark passing designs; a metric Phi does not score is shown but left out of the pass.
| Set | Thresholds | Source | Evidence |
|---|---|---|---|
| Dyno recommendedas of 29 Sept 2026 | pLDDT ≥ 80 · pTM ≥ 0.55 · ipTM ≥ 0.5 · ipSAE ≥ 0.35 · iPAE ≤ 10.85 Å · Binder RMSD ≤ 3.5 Å | Dyno, the public benchmark's default presetThresholds for comparing designs on the AlphaFold 2 metrics; they say nothing about the hit rate in the lab. | a working cut |
| Dyno recommended, relaxedas of 29 Sept 2026 | pLDDT ≥ 80 · pTM ≥ 0.45 · ipTM ≥ 0.5 · ipSAE ≥ 0.3 · iPAE ≤ 12.4 Å · Binder RMSD ≤ 4.5 Å | Dyno, the public benchmark's relaxed presetThe same metrics, loosened, for a screen that can take more designs. | a working cut |
| AF2 initial guess (RFdiffusion 3 benchmark) | iPAE ≤ 10 Å · Binder pLDDT ≥ 80 · Binder RMSD ≤ 1 Å | Dyno Psi-1 paper, in-silico benchmark: the AF2 initial-guess filter applied equally to RFdiffusion 3, BoltzGen and Psi-1Interface pAE under 10 Å, binder pLDDT over 80, binder RMSD under 1 Å; the paper also rejects TM over 0.6 to the natural partner, which Phi does not score. | stated in the method |
| Dyno Psi-1, pooled display | Binder RMSD ≤ 3.5 Å · Binder pLDDT ≥ 70 · ipSAE ≥ 0.1 | Dyno Psi-1 paper, yeast-display arm: the lenient filter for a pooled screen of thousands, then ranked by AF2 ipSAEBinder pLDDT over 0.7 on the paper's 0 to 1 scale (its main text prints the sign the other way; the appendix has it as a floor). | stated in the method |
| RFantibody, prospective | iPAE ≤ 10 Å · Binder RMSD ≤ 2 Å | Bennett et al. 2026, RFdiffusion antibody design: the filter used before orderingRF2 interface pAE under 10 read here as AF2 iPAE; Cα RMSD under 2 Å. | stated in the method |
| BindCraft default | iPAE ≤ 10.85 Å · ipTM ≥ 0.5 | BindCraft's default filters as Dyno runs themi_pAE under 0.35 on BindCraft's 0 to 1 scale is 10.85 Å on Phi's; the ipTM threshold is a binary call, not a ranking. | a working cut |
| AlphaProteo | No published thresholds | AlphaProteo paperZambaldi et al. 2024 describe an automated filter that chose 47 to 172 candidates per target and say its thresholds are not disclosed and vary by target. Numbers attributed to AlphaProteo elsewhere are other groups' reconstructions. |
| Evaluation status | Meaning |
|---|---|
| ok | Computed; the value is real. |
| missing_input | An input the metric needs was absent, so it was not evaluated. |
| not_applicable | The metric does not apply to this design or model. |
| failed | The computation ran and failed. |
| unsupported | The model does not produce what this metric needs (for example Chai-1 writes no PAE). |
Choose a filter set
A filter set is a named group of thresholds with its source attached. The built-in choices are:
- Dyno recommended, with the date of the recommendation.
- Published sets from the AF2 initial-guess filter used in the RFdiffusion 3 benchmark, Dyno Psi-1's pooled-display filter, RFantibody's prospective filter, and BindCraft's default.
- Yours, for a set you save under your own name.
The Binder lab uses one set at a time. In Plan, Thresholds shows the selected set, its source, and each cutoff. You can reset an edited value, save the result as a new set, or copy a link. In Review, the selected set appears beside the pass count. Switching sets recalculates the marks on the same designs. A generator's internal filters remain separate in that generator's settings.
How the source is described
Each set labels how its thresholds were obtained:
- Stated in the method means a paper gives the cutoff in its method.
- A measured outcome means the source reports the result.
- A working cut means the cutoff is in use without a published measurement.
If Phi does not score a metric yet, the set still shows its threshold but leaves that metric out of the pass. A method that does not publish its thresholds, such as AlphaProteo, is listed without invented values.
Share a set
A bench link carries the selected set and its thresholds. A colleague who opens the link sees the same marks and can save the set. Organization-wide sets and change history require Phi's recommendation registry and are not available yet.
Apply thresholds
A threshold is one metric and one cutoff. A design passes it when its value is on the better side.
Two rules apply a set of thresholds:
- All. Every threshold must be evaluated and pass. This is the API's
passes_allflag on every persisted design. A design missing any of the three metrics is not a pass. - Evaluated only. Only the thresholds a design was evaluated on count. This is how models with different metric coverage are compared on the same benchmark: BoltzGen reports no binder RMSD, so its designs are judged on the other thresholds and the RMSD cell reads not evaluated.
Whichever rule is in force, a missing measurement is never a pass.
Read evaluation status
Every evaluation the API stores carries a status. ok means the value is real. Anything else means the value is absent for a stated reason, and the page shows the reason, not a number.
Sequence scores
Design mode
ProteinMPNN returns three values for each designed sequence:
scoreis the average negative log-likelihood of the designed residues.global_scoreis the same calculation over every residue the model saw, including residues it kept.seq_recoveryis the share of designed residues that match the input.
Lower is better for score and global_score. No published source gives a selection threshold for these values, so the lab draws none. Phi does not treat recovery as evidence of function or use an ESM log-likelihood to pass or fail a design.
Refolding provides a separate structural check. ESMFold gives per-residue pLDDT for each design, and an overlay against the input gives Cα RMSD.
Residue scores
In probs mode, ProteinMPNN scores each position while holding the backbone and every other residue fixed. This leave-one-out calculation produces:
- Native fit, the log-probability assigned to the residue present. A value of 0 is a sure fit; −6 or below is very unlikely.
- Certainty, one minus the entropy of the model's 21-letter distribution over ln 21. A value of 1 means one clear choice; 0 means no preference.
The sequence score in this mode is the mean native fit. The lab calculates both values when a structure loads and shows them on the structure and under the sequence. Darker native-fit values mark residues the model considers a poor fit. Darker certainty values mark positions where the model has no clear preference.
Expert shows a value under every residue. Simple shows the value under the pointer. A native fit of −3.6 means the model assigned the residue present a probability of about e−3.6, or 3 in 100.
Read the propensity row
Select a residue to see ProteinMPNN's full distribution at that position. The twenty amino acids are ranked by probability, with values below one percent grouped together. Blue marks the working residue; after a substitution, a ring marks the native one.
A high probability for a non-native residue makes it a candidate to test, not a verdict, so no threshold is drawn. Selecting a letter changes the working sequence and recalculates the scores. Fold this sequence refolds the designed chain so its predicted confidence can be compared with the input.
Other inverse-folding models
- PottsMPNN returns the ProteinMPNN values plus Potts energy for each sequence. Lower Potts energy is better. It scores residues in the same form as ProteinMPNN.
- ESM-IF1 reports
scoreas the negated average log-likelihood over the designed chain andglobal_scoreover residues with coordinates. Lower is better for both. These are not ProteinMPNN values and should not be compared with them. - Proton-PottsMPNN reports Potts energy, selective energy, and the design's point on the sweep between them. Lower Potts energy is more stable; lower selective energy is more selective.
None of these models has a published selection threshold.
Checked 2026-10-05