Sticky vs. Specific: Cross-Checking Three Screens Against Each Other

Three full-library screens now exist — PTR1, TR, and CYP51 — three chemically unrelated pockets on two different Leishmania proteins. Conivaptan, nilotinib, rimegepant, and eltrombopag all showed up near the top of more than one of them. That's worth actually checking with numbers instead of just noting it in passing.

Step 1: The question

A drug that scores well against one pocket might genuinely fit that pocket's shape and chemistry — or it might just be a large, flexible molecule with lots of hydrogen-bond donors that Vina's scoring function rewards more or less everywhere it's placed, regardless of whether the fit means anything. With only one or two screens, there's no way to tell those two cases apart. With three, there is: a drug that's dramatically better on one target than its own baseline performance on the other two has a real reason to be there. A drug that's equally strong (or weak) on all three unrelated pockets is showing the signature of a generically "sticky" molecule instead.

Step 2: Method

All three screens docked the exact same 1,795 successfully-prepared drugs (the only thing that excludes a drug is a ligand-size limit, independent of which receptor it's docked against), so every drug already has three directly comparable scores sitting in the existing data — no new docking needed for this analysis.

• Data-quality filter first. A positive Vina affinity means the pose failed or clashed — it's not a real weak binder, it's noise. PTR1 had 2 of 1,795 positive-score drugs, CYP51 had 12, and TR had 33 — meaningfully more, plausibly because TR's larger, more sterically complex A+B dimer interface is a harder receptor for some bulkier ligands to fit into cleanly than PTR1's or CYP51's single-chain pockets. Excluded any drug with a positive score on any target, leaving 1,762 of 1,795 with three clean numbers to compare.
• Per-target z-scores. Standardized each target's affinity distribution (PTR1 mean −7.80±1.65, TR mean −6.53±1.46, CYP51 mean −7.97±1.81) so "how good is this score" is comparable across targets with different absolute scales.
• Specificity gap = (average z-score on the two weaker targets) − (z-score on the best target). A large positive gap is real evidence a drug's best result isn't just its baseline performance everywhere — something about that specific pocket is doing extra work. A gap near zero means the drug is equally strong (or weak) against all three unrelated pockets, the signature of a molecule Vina just likes in general.
• Ligand efficiency (LE) = |best affinity| ÷ heavy-atom count (counted via RDKit from each drug's SMILES) — a standard med-chem sanity check, since Vina's raw score is known to correlate with molecular size. A big molecule racking up a good score mostly by being big should show unremarkable LE even with an impressive raw number; a small molecule matching a pocket precisely shows high LE.

Step 3: Every drug discussed across the three screens so far

Bold z-scores mark each drug's best target. Read thresholds: gap > 1.2 = SPECIFIC, gap < 0.6 = STICKY, in between = mixed.

Step 4: What this changes

8 of 25 are confirmed generically sticky — conivaptan, tepotinib, tolvaptan, nilotinib, eltrombopag, olaparib, dihydroergotamine, entrectinib. Consistently strong across all three chemically unrelated pockets, essentially the same z-score no matter which target you look at. Their good raw scores are real, but the evidence points to "Vina likes this molecule in general," not "this molecule fits this specific pocket." This confirms the informal suspicion flagged after the CYP51 screen (conivaptan showing up a third time, nilotinib a second) — it wasn't a fluke, it's the majority pattern for repeat hits.

4 of 25 are genuinely pocket-specific — and this is the part worth paying real attention to: they're also exactly the four drugs with the strongest independent literature corroboration found across all three screens. Digitoxin and irinotecan both have confirmed activity against L. infantum itself (see the CYP51 post), and alectinib was PTR1's original #1 hit. That's a real, independent cross-check on this whole method — it's picking out the same drugs the literature separately flags as most biologically plausible, not just re-describing the docking scores in different language. Bexarotene has no literature precedent yet, but posts the single best ligand efficiency in the entire table (0.438, from only 26 heavy atoms) — a small molecule fitting the pocket precisely rather than a big one racking up contacts by sheer size — worth a second look for that reason alone.

Step 5: A correction to the TR post's framing

The TR post called ergotamine the project's most promising lead, and that framing needs walking back a little: ergotamine is "mixed, leaning sticky" (gap 0.97), not cleanly pocket-specific. It's a strong binder on all three targets, just clearly strongest on TR — real signal, but weaker specificity evidence than "novel, pocket-matched scaffold" implied.

This is actually unsurprising once you think about the chemistry: ergot alkaloids are famous for promiscuous polypharmacology even against human receptors — dopamine, serotonin, and adrenergic receptors are all real, well-documented ergotamine targets. A scaffold already known for hitting many unrelated binding sites in humans has no particular reason to suddenly become pocket-selective against a novel parasite target instead. This doesn't disqualify ergotamine — it's still the cheapest, most cross-run-consistent hit the project has found, and promiscuous compounds have a real (if less mechanistically tidy) history as antiparasitics. The honest story going forward is "consistently strong, likely non-selective binder, still worth the cost argument" rather than "novel, pocket-matched scaffold."

Ligand efficiency, for what it's worth, didn't cleanly separate the two groups on its own — sticky and specific drugs both span roughly 0.2–0.35 LE, aside from the standout cases (bexarotene's 0.438; digitoxin/irinotecan in the 0.24–0.29 range despite being fairly large molecules). The specificity gap is doing essentially all of the discriminating work here; LE is a useful secondary signal, not a standalone filter.

What's next

This is now a standing step for any future screen on this project: before getting excited about a top-10 hit, check its score against the other targets already run. A drug that's merely "consistently good everywhere" is a weaker novel-mechanism claim than a drug that's dramatically better on the new target specifically — and now there's a real, literature-cross-checked method for telling the difference instead of just a hunch. None of this is proof of mechanism — a high specificity gap is evidence consistent with genuine pocket complementarity, not confirmation of it, and a low gap doesn't prove a drug has no real antiparasitic activity via some other mechanism, only that this particular signal can't distinguish it from a generic strong binder.