brugada.net · What died

Updated 16 July 2026

The bodies are the evidence.

A screen that never reports a failure isn’t careful. It just isn’t testing anything.

This is the part of a research project nobody publishes, which is exactly why it’s here. Every molecule below looked promising enough to spend real computer time on, and every one of them failed. The most important one is the molecule this entire project was built around.

Agmatine

The hero. Killed by its own numbers.

Looked like a clear winner in the cheap test. Put in a proper simulated membrane, three times over, it lets go. Then the free-energy calculation finished it off: agmatine would rather not be in the clip at all.

Metformin

Gripped for 4 ns. Left.

A common approved drug that looked like a hit on a short read. The full test retracted it: it holds for about four nanoseconds, then the grip blows apart to roughly 21 Å.

Famotidine

Held 14 ns. Left.

The late escape — exactly what the 20-nanosecond stage was built to catch. A molecule that grips and then peels is worse than one that never gripped, because it survives the cheap test and wastes the expensive one.

4-guanidinobenzamidine

Designed for the job. 1 of 3.

Not found — designed. Rigid, two-handed, everything the theory asked for, and the fast tool loved it. It engages and does not stay. Being the right shape isn’t enough.

Zoniporide

New. Dead this week.

A real drug, tested in the run going on right now. Ends the test 18 Å from the target with an occupancy of 0.002. That’s not a near miss; that’s a no.

Mexiletine

Right idea. Wrong pocket.

The drug that proves this class of rescue can work — it fixed a different variant of this same protein 93-fold. It doesn’t bind this site. Precedent isn’t a solution.

The one that mattered

This project falsified its own best idea, and published the graph.

Agmatine was the lead. Everything was built around it. Then someone ran the expensive calculation that measures how much a molecule wants to be somewhere — and the answer came back: not there.

Binding free energy of agmatine along the clip coordinate Umbrella-sampling potential of mean force for agmatine against the distance from its guanidinium to D84. The minimum is -1.94 kcal/mol at 8.6 angstroms — a separated-contact state, not the clip. The clip GO band near 3.3 angstroms sits at +7.55 kcal/mol, up the repulsive wall. The well is only about 3 kT deep. Thermodynamically, agmatine prefers not to be in the clip. FREE ENERGY OF AGMATINE ALONG THE CLIP COORDINATE · UMBRELLA SAMPLING · 22 WINDOWS · 26,400 SAMPLES The lead that did not survive. Lower = more favourable. Shaded band = ±1 standard error. -2 +0 +2 +4 +6 +8 4 6 8 10 12 14 distance from agmatine's guanidinium to D84 (Å) ΔG (kcal/mol) −1.94 kcal/mol at 8.6 Å a separated state — not the clip. Only ~3 kT deep. +7.55 kcal/mol — where the clip is the grip it was nominated for sits up the repulsive wall 1 kT = 0.616 kcal/mol
II

Lower on this graph means “more comfortable”. The pink dot on the left is the spot agmatine was supposed to grip — and it’s at the top of a hill. The molecule’s actual preferred position is the dip in the middle, 8.6 Å away, not touching. And that dip is shallow enough that ordinary room-temperature jiggling knocks it out.

What the graph is actually measuring

This is a potential of mean force from umbrella sampling: 22 windows, 26,400 samples. It maps the free energy of the system against the distance between agmatine’s guanidinium and D84.

The minimum is −1.94 kcal/mol at 8.6 Å — a separated-contact state. The clip GO band (~3.3 Å) sits at +7.55 kcal/mol, up the repulsive wall. The well is ~3 kT deep, i.e. thermal motion clears it.

This is also a correction: the result file originally labelled that minimum a “favorable clip well”, which was wrong — position-blind logic called any deep well a clip well regardless of where it was. Relabelled; numbers untouched. The corrections page →

Results/VAST_HARVEST/pmf/pmf_curve.csv · PMF_RESULT.json

That is the only free-energy calculation this project has ever completed, and it killed the project’s own lead. It has never been run on any of the current candidates. If you want one number for how far this has left to go, that’s the one.

onwards.