PXDesign

PXDesign is a no-install online AlphaFold2-scored binder design tool you can run through tools.ranomics.com on a dedicated GPU. The same pipeline Ranomics runs for its own wet-lab campaigns: every candidate comes back already re-folded against your target, carrying its own confidence score for the contact rather than a number borrowed from the generator.

Heads up. Long-running job. PXDesign pilot runs 8 to 25 min on A100-80GB. That is every pilot run on record, measured at 2 to 25 designs against targets up to about 420 residues; more designs cost more but do not add time in step. Past that, nothing has been measured, so allow longer. Stay on this page or return via /jobs to check status; results auto-refresh when done.

Design inputs

Binder length (residues)

Target

Target PDB

Upload the target structure as .pdb, .cif, or .mmcif.

Target chain

One chain (A), or several for an oligomeric target (A,B). Every named chain reaches the model; designing against one protomer of a dimer aims at half the epitope.

Hotspot residues

Residues the binder should contact, in original PDB numbering. A plain number is read as your first target chain; prefix the chain to name another (A296,B264). Click residues in the 3D viewer below to toggle them in and out of this list.

Not sure which residues? Score your target's surface with Epitope Scout first.

3D residue picker

Surface

Upload a target PDB above to enable click-to-pick. Typed entry in the Hotspot residues field above also works.

Designs

Number of designs

One run does up to 24 designs. No fixed ceiling: larger requests fan out automatically as a wallet-bounded campaign across GPUs.

Runs against your own target. Upload PDB, pick hotspots, set binder length. Results emailed (8 to 25 min on A100-80GB).

Campaign

Group related runs on your jobs page. Leave blank to file under Uncategorized.

Estimated cost $0.00 based on parameters and historical compute time

Billed by actual GPU consumed, up to the estimate. Surplus returns to your wallet automatically. Cancellation does not refund GPU time already used.

About PXDesign

What it is

Designs binders against your target and then scores every one of them with AlphaFold2 run against that same target, so nothing reaches you on a cheaper stand-in number. The scoring uses AlphaFold2's initial-guess mode, which is far quicker than a full multimer prediction and is what makes per-candidate scoring affordable at all. This is the pipeline Ranomics runs for its own wet-lab campaigns. PXDesign is in-house; the initial-guess method is Bennett et al., Nature Communications 2023.

When to use

You want an AlphaFold2 confidence number against your actual target on every candidate, not on a filtered subset.
You want that scoring faster than a full AlphaFold2 multimer run on each design.
You want the same pipeline Ranomics runs for its own wet-lab campaigns.

Prerequisites

Target structure (.pdb / .cif).
Chain ID of the target.
At least one hotspot residue.

Input glossary

Hotspot residues. Comma-separated target-chain residues defining the epitope. Click in the 3D viewer to toggle.
Binder length. Residue count for the generated binder. ~40 residues is the validated default; PD-L1 published binders are in this range.
Number of designs. How many candidates to score. Cost rises with the count. Runtime rises far more slowly and not in step with it: two designs took 8.4 min, five took 7.7, and twenty-five took 23.0. Past twenty-five nothing has been measured, so allow more time.

Runtime

Preset	Typical min
pilot	8 to 25 min, measured at 2 to 25 designs

Ranked candidates with ipTM, pLDDT, pAE, and PDBs downloadable from a run of your own. Target ipTM ≥ 0.75 for a tractable epitope.

References

Bennett, N. R., Coventry, B., Goreshnik, I., et al. "Improving de novo protein binder design with deep learning." Nature Communications 14, 2625 (2023). Ranomics in-house pipeline; scoring stage uses AF2 Initial Guess.

Starter pilot: 4 designs

See whether your target and the face you picked give designs with usable AF2 confidence, before scaling.

What you need

A structure file for your target (.pdb or .cif), the chain ID, and at least one residue on the face you want bound.

Hotspot residues are the numbered residues on the patch of your target you want the binder to sit on, written in your PDB’s own numbering — this tool will not start without at least one. If you do not know yours yet, score your target’s surface with Epitope Scout first — it is free, and its results hand the target and the residues back into this form.

What it costs

About $12.59, and about 8 to 25 min of GPU time. That is an estimate from these exact settings — you are billed for the compute the job actually uses, never more than the estimate.

Then what

If any of the four comes back with good interface confidence, clone the run and raise the design count.

Load these settings →

A run we actually did

The target

A two-chain human secreted protein, about 210 residues per chain, from a solved crystal structure. Six hotspot residues, three on each chain, across the two-fold interface where the chains meet.

The binding site sits on a symmetry axis, so a binder has to reach both chains at once rather than settle on either one. That is a harder ask than a single flat face, and it is why so much of this round fails in the specific way shown below.

What we put in the form

Target PDB — the two-chain crystal structure: Both chains staged together. Handing over one chain would let the model design against half a site that does not exist on its own.
Hotspot residues — 6 residues, 3 per chain: The face we wanted engaged, given symmetrically so neither chain reads as the whole target.
Binder length (residues) — 63: One of four calls we made at 50, 57, 63 and 70. The pilot tier draws a fresh seed per call, so fanning out over lengths buys seed diversity and length coverage from four calls.
Number of designs — 25: Enough to see the shape of the failure, cheap enough to throw away. 100 across the four calls.

What came back

25 designs, of which one reaches the bar this site applies — ipTM 0.75 and pLDDT 80, the numbers in the column tooltips. The best scored ipTM 0.88 at pLDDT 88, with the re-folded complex landing 1.64 Å from where the generator put it and 577 Å² of surface buried against the target. The other 23 are the point of showing you this. The median pLDDT in this call is 91 — high — while the median ipTM is 0.14. 21 of the 25 designs fold beautifully and do not bind anything. All four calls looked like this; 8 of the full 100 cleared the pipeline’s own filter at the time. That count is the PIPELINE’s, recorded when the round ran. The table above is scored against this site’s bar every time you load it, which is why the two numbers need not agree: one is a record of what a container decided in 2026, the other is a comparison made just now. The second design the pipeline passed sits at pLDDT 78, under the 80 bar; it cleared because PXDesign’s own summary said so, not because it met a threshold.

Candidates (25)

#	ipTM ?	pLDDT ?	pAE (Å) ?	vs. quality bar ?	3D	Structure
1 Top	0.880	88.0	6.28	Meets ipTM 0.75 and pLDDT 80	—	—

2	0.770	78.0	9.83	pLDDT 78.0, below 80	—	—

3	0.170	90.0	27.36	ipTM 0.170, below 0.75	—	—

4	0.150	90.0	27.82	ipTM 0.150, below 0.75	—	—

5	0.140	91.0	28.00	ipTM 0.140, below 0.75	—	—

6	0.140	91.0	27.78	ipTM 0.140, below 0.75	—	—

7	0.140	80.0	28.17	ipTM 0.140, below 0.75	—	—

8	0.140	93.0	28.04	ipTM 0.140, below 0.75	—	—

9	0.140	91.0	27.89	ipTM 0.140, below 0.75	—	—

10	0.140	92.0	28.00	ipTM 0.140, below 0.75	—	—

11	0.140	92.0	28.12	ipTM 0.140, below 0.75	—	—

12	0.140	88.0	28.26	ipTM 0.140, below 0.75	—	—

13	0.140	93.0	28.04	ipTM 0.140, below 0.75	—	—

14	0.140	68.0	28.36	ipTM 0.140, below 0.75; pLDDT 68.0, below 80	—	—

15	0.140	91.0	28.06	ipTM 0.140, below 0.75	—	—

16	0.140	93.0	28.07	ipTM 0.140, below 0.75	—	—

17	0.140	88.0	28.31	ipTM 0.140, below 0.75	—	—

18	0.130	91.0	28.18	ipTM 0.130, below 0.75	—	—

19	0.130	86.0	28.17	ipTM 0.130, below 0.75	—	—

20	0.130	85.0	28.24	ipTM 0.130, below 0.75	—	—

21	0.130	92.0	28.21	ipTM 0.130, below 0.75	—	—

22	0.130	91.0	28.26	ipTM 0.130, below 0.75	—	—

23	0.130	92.0	28.17	ipTM 0.130, below 0.75	—	—

24	0.130	92.0	28.19	ipTM 0.130, below 0.75	—	—

25	0.130	93.0	28.24	ipTM 0.130, below 0.75	—	—

Validating in the lab is optional Designs from your own run are yours to download and take anywhere. If you do want them made and measured, Ranomics runs yeast display, mammalian display, BLI and DMS on the same data loop.

Binder Pilot AI Binder Sprint

How to read that

Those two columns answer different questions and this round is what it looks like when you confuse them. pLDDT is the model's confidence in the shape of the binder considered alone; ipTM is its confidence that the binder and the target form the complex you asked for. A design can be a textbook helical bundle and dock nowhere near the site, and 21 of these are exactly that. Sort by ipTM, never by pLDDT. The pipeline’s own filter, the one behind that 8-of-100, is the tool re-folding each design from scratch and checking it lands where the generator claimed: ipTM at or above 0.50 and the re-folded complex within 3.0 Å. That second half is what bites — across the round the median complex misses by 13.7 Å, while all 8 survivors land inside 2.3 Å. A design that clears the first and fails the second has folded correctly in the wrong place.

What happened next

Kept the 8 and re-ran at longer binder lengths. Passes by length here were 0, 2, 2 and 4 out of 25 at 50, 57, 63 and 70 residues, which is 25 per bin — suggestive, nowhere near enough to call a length effect, and the reason the next round was a scale-up rather than a conclusion. If your own pilot returns a page of high pLDDT and low ipTM, that is not a broken run: it is this result, and the answer is more designs or a different site, not a different setting.

Recorded from that one run, which cost $2.41, and took 23 minutes. That is what it did, not a quote for yours — the estimate on the form is the live number for your own inputs.

What good looks like

Each tool reports a subset of these depending on whether it does design, sequence recovery, or structure prediction.

ipTM: Predicted confidence in the contact between two chains, on a 0 to 1 scale. Higher is better: 0.65 to 0.75 depending on the tool. Individual tools set their own pass bar a little either side of that — the guide for the tool you ran states its one. On a multi-chain target the number may cover the target's own chain–chain interface as well as the binder's.
pLDDT: Per-residue confidence in the predicted fold. Higher means the model is more sure of that part of the structure.
i_pAE and pAE: Predicted alignment error, at the interface (i_pAE) or across the whole structure (pAE). Lower is better.
ProteinMPNN recovery: Fraction of native residues recovered when ProteinMPNN redesigns a known sequence on its native backbone — the folded shape on its own, before a sequence is chosen for it. Higher is better: > 0.4 usable; > 0.6 excellent.

Related tools on Ranomics

If you are picking between PXDesign and a sibling algorithm, these run on the same hub against the same target.

Run RFdiffusion online: You have a target structure and a patch of its surface you want gripped, and you want brand-new binders of whatever shape works — the general-purpose starting point for de novo design. Every candidate comes back with a real AlphaFold2 confidence score against your own target. For antibody or nanobody formats use RFantibody or IgGM instead.
Run BindCraft online: You have a target structure and know roughly which patch of its surface you want gripped, and you want brand-new mini-proteins of 50 to 150 residues built to grip it. Every candidate is refolded and filtered before you see it, so what comes back is already a shortlist.
Run BoltzGen online: You have a target and have not settled on what shape the binder should be. One model here aims mini-proteins, nanobodies, antibodies or peptides at the same site, so you can compare formats instead of guessing. The only design tool here that handles sugars and modified residues on the target natively.