GPCR Foundation Model

Rank compounds against one receptor

Ligand A → sequence → ligand B. Pick a receptor, supply two or more compounds, and the model orders them for that receptor.

Nothing is scored on its own and then subtracted. The forest is handed the whole comparison as one row, ligand A, the sequence, ligand B, in that precise order, and returns the probability that ligand A is the more potent of the two. Every pair of the compounds you supply is put to it that way, in both ligand orders, and the two answers are averaged.

A worked potency comparison. Two ligands measured at the dopamine D2 receptor, a substituted benzamide at pKi 11.52 and an aryloxy-alkylamine at pKi 3.00. Each ligand becomes a 1,024-bit Morgan count fingerprint plus 14 descriptors; the receptor sequence becomes 480 numbers through ESM2. All three blocks enter a random forest as a single row, which returns which ligand binds tighter.
The potency comparison, end to end. Both ligands and the receptor enter as one row, in the order ligand A, sequence, ligand B. Nothing is scored on its own and subtracted. The two compounds shown are real training measurements at the dopamine D2 receptor, 8.5 log units apart.
A, measured pKi 11.52 at DRD2
CCN1CCC[C@H]1CNC(=O)c1cc(Br)cc(OC)c1OC
B, measured pKi 3.00 at DRD2
COc1cccc(CCc2ccccc2OCCN2CCc3ccccc3C2)c1
Paste both into the box above with DRD2 selected to reproduce this comparison.
Step 1

Receptor

Receptors the model was fitted on are grouped first, with their held-out accuracy and comparison count. Names below that group are recognised but were not in the training set, and a result for one of them is unvalidated.

A sequence the model was not trained on is scored and flagged. It is not covered by any accuracy figure on this site.

Step 2

Compounds

Two at minimum, 2 at most per run, drawn and pasted combined. Every pair is scored, so n compounds is n(n−1)/2 comparisons.

You do not have to draw it. To paste a SMILES string use the editor's Open Structure button, the folder icon at the top left, and choose Paste from clipboard. It accepts SMILES and will draw the molecule for you. Then press the button below to read it back out.
You do not have to draw it. To paste a SMILES string use the editor's Open Structure button, the folder icon at the top left, and choose Paste from clipboard. It accepts SMILES and will draw the molecule for you. Then press the button below to read it back out.

Drawing structures never needs an email. Pasting or uploading a list of compounds does, and the full report is emailed back to you.

Unlocks the paste box and the file upload below ↓

We send the full results there, and let you know when the models change.

 up to 2 structures per run
Step 3

How to read the confidence

Prediction strength is the larger of the two output probabilities, so it runs from 0.5, a coin flip, to 1.0. These are the accuracies measured for each band on the held-out set, and they are what the colours in the result mean.

StrengthRight this oftenShare of comparisons
0.80 and above89.2%1.3%
0.70 to 0.8086.1%5.3%
0.60 to 0.7076.7%19.6%
0.50 to 0.6058.5%73.5%

Do not read past 0.80. The curve turns over there, falling from 0.887 to 0.869 on 1,537 comparisons, so this page does not offer a higher cutoff.

The cutoff does not change the order. Every comparison counts toward the ranking; the cutoff decides which ones are reported as firm.

Step 4

Run

Accuracy on this page is measured over comparisons between receptors the model was fitted on, 182 of them with a held-out score. It says nothing about a receptor the model has never seen.