Luminary Studies

Where a drug has to get to, and whether it can get there

Two pieces of one platform. The first maps thirteen disease environments — what is actually in them, how they are built, and what stands between a drug in the bloodstream and the cell it is meant to reach. The second designs a drug built around timing rather than strength, and tests it against those environments. Everything here is computation on published measurements. Nothing was measured by us, and no molecule described here has been made.

11 October 2026 · in active development · computational work only

the map · 13 environments Microenvironment Atlas Colorectal, gastric, lung, prostate, pancreatic, breast, liver, ovarian, glioblastoma and melanoma — plus ALS, frontotemporal dementia and Alzheimer's. Each one drawn with its own published numbers, each with a treatment timeline you can play through. Every site is a different problem: a wall of pressure in one, a 20-nanometer gap in another, a target hidden on the wrong face of a cell in a third. the design · one molecule Selectivity Is a Statement About Time A drug that tells a cancer cell from a healthy one by how long it holds on, not how tightly it grips. The full derivation, with every number traced to its source and a section listing every assumption we could not close.

why these are one thing

A molecule is only as good as the place it has to work in. The atlas says what the place is made of; the design says what a molecule would have to do there. The goal is to run a drug against any site on the map — its chemistry, its structure, the components it would meet, the reactions to expect — and get back an honest answer about whether it reaches its target in that tissue, because the answer genuinely differs from one site to the next.

See it run

Two of these are not documents. They are drawn to scale, they play, and everything labelled in them can be clicked for what it is — or for a plain statement that nobody has measured it.

play · thirteen tissues Where the drug has to get to Pick a disease, press play the treatment span, and watch a dose cross that particular tissue: the pressure, the seal, the wall, and where the target turns out to be sitting. Pancreatic and Alzheimer's are the two to see first — they fail for opposite reasons. play · three scales Observations at the synapse Zoom from 150 µm of tissue to the 450 nm gap between two touching cells. Nine conditions decide whether the cell dies; try runs to apoptosis, then any of the six ways it fails, and watch which gate stops it. run · live workspace Put a drug into the microenvironment Load a drug or gene context, change oxygen, nutrients, immune pressure and exposure, then inspect the modeled state trajectory. Supported drug mechanisms change the cell state; everything else stays explicitly context-only.

Start from zero

No background needed. Four steps and the main figure will make sense.

A drug with two hands

Your immune system already contains cells that kill cancer — T cells. The problem is that they often fail to notice the tumor. So one class of drug is built as a molecule with two hands: one grabs the cancer cell, the other grabs a T cell, and holding the two together forces the introduction. The T cell does the killing; the drug just makes the handshake happen.

And the reason it hurts people

Healthy cells are not different enough. Whatever marker the drug grabs on a tumor cell usually appears on some normal cell too, in smaller amounts. The drug grabs those as well, the T cell kills those too, and that is where the side effects come from — real ones, serious enough to have ended drug programs and, in one case, to have killed a patient within five days.

So the hard problem is not killing. It is telling the two cells apart.

Our idea: require two markers, not one

Give the cancer-grabbing side two separate small hands instead of one big one, each looking for a different marker. A tumor cell carries both, so both hands catch and the drug stays put. A healthy cell carries only one, so only one hand catches — and one hand is not enough to hold on. It slips off within seconds.

This changes the question being asked of the drug. It is no longer how strongly does it grip? — it is how long does it stay? That is why this work is about time.

The experiment, and what "the read" means

To find out whether any of this works, a lab does something simple. Put cells in a dish. Add the drug and some T cells. Wait. Then count how many cancer cells died.

The waiting time matters enormously, and in the published experiments this design is based on it is six hours. We call that moment the read — when the dish is scored. Anything that has not happened by then does not appear in the result.

One more word. The drug does not grab once and stay — it grabs and releases, continuously, until the number held at any instant stops changing. That balance point is the population having settled. How long a population takes to settle is the quantity that decides whether a six-hour experiment can tell two kinds of cell apart — because if both have already settled before the read, the clock has nothing left to measure.

The idea in one picture

Now the figure says something. One time axis, ten seconds to a month. Three bands, and where each sits relative to the read is the whole argument.

Read it as two bands and a line. Gray is the healthy cell: it has let go and gone quiet within the hour, long before anyone looks. Green is the cancer cell: still being held when the dish is counted at six hours. The line is the moment of counting. Because the gray band finishes before the line and the green band crosses it, a six-hour experiment sees killing on one and not the other — and that gap, not any difference in grip strength, is the mechanism.

On the two weakest versions. Six variants of the molecule have been published, identical except for how tightly the second hand holds — a 138-fold range from firmest to loosest. The looser that hand, the sooner the cancer-cell population settles down. The two loosest settle at 5.3 and 5.5 hours, just inside the six-hour read, so the clock has less to work with — and in the original authors' own measurements those same two were the worst at distinguishing cancer cells from healthy ones. The timing explanation and their data point the same way.

What is on the platform

Four pieces, and what each one is for.

Microenvironment Atlas Thirteen disease environments drawn to scale with their own published measurements. Choose one, press play, and watch a treatment unfold through it: where the drug arrives, what blocks it, where the target turns out to be sitting, and what actually happened in the clinic. Click anything in the drawing to inspect it. Selectivity Is a Statement About Time The full technical paper behind the two-handed design — the model, the derivation, 35 tables of results, and a limitations section naming every assumption that remains open. This is the detailed version of the four steps above. Observations at the Synapse A guided zoom from a sheet of tissue 150 microns across down to the 450-nanometer gap between two touching cells, drawn to scale at every step. Nine things have to go right in sequence for a cell to be killed; this shows each of them, and names every player involved. Why Here and Not There Six real drugs of this class that went into patients, and what stopped each one — mapped onto the nine steps so you can see which step failed in which trial.

What we have established

Results that hold, each one built as a calculation and then tested against published measurements from outside this program.

result	why it matters
Selectivity scales directly with the second hand's grip — a measured slope of +1.02 (95% interval +0.92 to +1.11), where the competing explanation predicts 0	The two accounts do not differ by a margin, they differ by a whole power. That makes them separable by a single experiment.
The holding advantage comes from the hand nobody has modified — the other hand's grip strength cancels out of the arithmetic entirely	Verified against the exact solution to within 0.3%. Both published series varied the hand that cannot change this, so the prediction is untested, cheap, and falsifiable.
Required dose × waiting time is the fixed quantity, not dose alone — 241 in the natural units	Fixed in advance by our own arithmetic, then checked against an unrelated group's published data, where it held. The first prediction from this work to survive outside contact.
The floor on how little target a cell can carry is a counting limit, not a transport one — each bridge uses up one marker, so the drug bound can never exceed the markers present	Contains no rate constants at all. Run backwards it places a hard ceiling on other drugs' requirements, which is how we tested our numbers against two published trials without needing their internal data.
There is no single universal threshold of how much marker a cell needs — one published drug needs about 10,000 copies, another kills a cell carrying 875	Four independent routes to the same conclusion. It is why the threshold has to be computed per molecule, which is what this program does.
Only 3 of 37 targets surveyed across thirteen diseases are physically reachable by an antibody	The atlas's central finding, and the reason the map and the molecule belong in one platform. Reaching the target is a harder problem than choosing it.

How this work is done

Nothing here was measured by us. It is arithmetic on rate constants and measurements other laboratories have published, each one read for its numbers rather than its abstract. No molecule described here has been synthesized.
Every idea gets a prior-art search attached to it in the same session. Nothing leaves this program described as new until someone has gone looking for whether it already exists.
A measurement we have inspected and a figure we have been told are recorded differently, even when they agree. A result quoted from a table we have not seen is the weakest thing to build on, and is labeled as such.
Where a number does not exist, the gap is published instead of filled. The atlas names 58 of them — including several that would change our own conclusions if they came back the wrong way.

A drug that binds its target more tightly is not the same as a drug that reaches it. Thirteen environments, and the distance between those two sentences is different in every one.

Luminary Studies · computational work on targeted cancer and neurodegenerative therapeutics. The molecule under design is a three-part format: one hand to anchor, one to confirm the second marker, one to recruit a T cell. All figures are drawn from published literature; measurement methods are stated alongside values and are never mixed on a shared axis. This platform is in active development and its contents change as the work does.

The Engager ProgramLuminary Studies