Compare two targets for one compound

Sequence A → ligand → sequence B. Pick two targets, supply one or more compounds, and the model says at which target each one is more potent.

"More potent" spans binding affinity and functional potency together: Ki and Kd are binding constants, EC50 and Kb are functional readings, and IC50 is an inhibition concentration. A comparison is always between two readings of the same kind, so the answer is an ordering of like with like rather than a claim about binding in particular.

The layout is the mirror of the potency model. The ligand sits in the middle and the two targets are the things being compared, so the question the forest answers is which target wins, not how tightly either binds.

A worked selectivity comparison. One ligand, a 4-anilidopiperidine, measured at two opioid receptors: pKi 10.89 at the mu target and 6.91 at the kappa target. Each target sequence becomes 480 numbers through ESM2 and the ligand becomes a 1,024-bit Morgan count fingerprint plus 14 descriptors. All three blocks enter a random forest as a single row, which returns which target the compound prefers. — **The selectivity comparison, end to end.** The ligand sits in the middle and the two targets are what is being compared, so the row reads sequence A, ligand, sequence B. The compound shown is a real training measurement, roughly 4 log units mu-selective over kappa.

**The ligand**, measured pK_i 10.89 at OPRM1 and 6.91 at OPRK1
`CCC(=O)N(c1ccccc1)[C@H]1CCN(C[C@H](O)c2ccccc2)C[C@H]1C`

**A kappa-selective counter-example**, measured pK_i 6.27 at OPRM1 and 10.21 at OPRK1
`Oc1cccc(CCN(CCc2ccccc2)CC2CCCCC2)c1`

Either structure pastes into the box above with OPRM1 and OPRK1 selected. They rank opposite ways.

The model is sequence, ligand, sequence, and the steps below follow that order: the first target, then the compound, then the second target.

Step 1

Sequence A, the first target

Target A

Targets the model was fitted on are grouped first with their held-out accuracy. Selectivity accuracy varies by target: across the 104 targets carrying at least 50 held-out comparisons the median is 0.82 and none fall below 0.60. Closely related targets are the hard ones.

Step 2

Ligand

The compound put to both targets.

Ligand: draw or paste

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. Then press the button below to read the structure back out.

Step 3

Sequence B, the second target

Target B

Or screen a list against this pair

Instead of the editor, give a list. Each compound is put to both targets. 2 at most per run.

Screen a larger list: enter your email

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

Unlocks the paste box and the file upload below ↓

Results go to that address, and we use it to say when the models change.

Or paste a SMILES list, one per line, optional name after a space or tab

up to 2 structures per run

Step 4

How to read the confidence

Strength runs from 0.5 to 1.0, as on the potency page. What a given strength buys is not the same on the two models, so use the numbers below rather than carrying a threshold across models.

Strength	Right this often	Share of comparisons
0.90 and above	99.0%	17.7%
0.80 to 0.90	94.2%	18.8%
0.70 to 0.80	86.5%	18.7%
0.60 to 0.70	72.3%	21.8%
0.50 to 0.60	57.6%	23.0%

Only report calls at or above

Step 5

Run

Two targets a compound cannot tell apart are two targets the model cannot tell apart either. The muscarinic targets M1 through M5 are the closest case on this site and score between 0.75 and 0.78. Opioid targets reach 0.90.