Luminary Health · microenvironment atlas
Thirteen disease environments, drawn from published tissue measurements and explicitly identified assumptions. Each environment has an evidence inventory, with measured dimensions where available and gaps left visible. The disease and treatment chapters explain proposed sequences; they are not validated predictions of an individual patient's response. Colorectal opens first and is drawn in the most detail.
The inventory supports model construction: an entry tells us what to represent and which evidence to consult. It does not mean that the component already has executable reactions or measured rates. Before using a scene quantitatively, check its Evidence gaps and simulation readiness below. Research findings are connected there to the selected environment.
Three things to do here. Click any component — the numbered ones are drawn above, and beneath them is the full inventory this environment's evidence contains, grouped by where it sits: 37 to 70 component families per environment, each carrying its own citations. Play the two chapters. How it forms covers sixty years, most of which have nothing to see; what a dose does covers forty-eight hours in the same tissue. Every step is listed with its own time, so the shape of it is visible before you press anything, and you can hold on any step as long as you want to read it. Follow a medicine from the blood to its target — a real engager, a plain antibody, a T cell, a cytotoxic — to wherever it stops, and then see that contact drawn at the scale it happens on.
11 October 2026 · nothing here was measured by us · every figure is published or marked NOT MEASURED
Run this environment in the live workspace →One of the thirteen has been taken further: the colorectal module splits it into four compartments — lumen, crypt, invasive tumour, liver metastasis — and runs each one twice, as the disease left alone and as the disease treated.
—
—
The components in this environment. The numbered ones are drawn above; the rest are present in the tissue and not drawn. Click any of them to read what has been measured about it.
Each component carries its own measured dimension, or says plainly that nobody has measured it.
Choose one and watch where it gets to. Each step says what is left of it, what stopped it, and where the honest answer is that nobody has measured that step.
Can an antibody physically touch it? Can a T cell?
The distance a molecule still has to cross once it leaves the blood.
The distance is one question; the size of the opening is the other.
Measurements we still need, and what the existing work can support.
An antibody arrives from the blood, so it meets the basolateral face. A T cell infiltrates, so it does not. That single difference decides which of these thirteen environments a gated engager can and cannot help in.
Bringing a molecule rather than reading about a tissue? The Molecule Library holds fifty-two of them, draws how each is built, and orders all thirteen environments by how much of it reaches the target.
Wanting to change one rather than look one up? The Molecule Design Lab takes a real structure apart and puts a group where you point, or alters one stage of a mechanism, and says in both cases what the change settles and what it leaves exactly as unknown as it was.
atlas state · 19 October 2026
The atlas behind this drawing has grown. It now audits twenty disease environments across 110 component families, every family profiled, against 482 records — 326 of them with their numbers extracted rather than just their abstracts read. The drawing here still covers thirteen; the other seven are written up and not yet drawn.
What that buys for a candidate molecule is narrower than a six-stage path suggests, and the honest version is worth more than the optimistic one. For an antibody arriving from the blood into a solid tumour, the vessel wall alone puts the concentration reaching the tissue surface one hundred to one thousand times below plasma. How much less reaches a cell a hundred micrometres further in is not known: the pore geometry that would set it is unmeasured in all twenty environments. The only decay length held at that scale belongs to a small molecule — doxorubicin halving every 40–50 µm, with viable cells sitting 90–140 µm from the nearest vessel — and it does not transfer to an antibody.
Two things have since been pinned down about that sentence. The hundred- to thousandfold drop is antibody-sized traffic and not the wall: measured across carriers from 3.3 kDa to 2 MDa, permeability spans ninety-fold, so a small macromolecule crosses about fifty times faster than IgG and a two-megadalton carrier about twice as slowly. And the distance on the far side is no longer an assumption in one environment. Breast has a measured vessel density, and building the field from it puts the middle of the tissue 41 µm from the nearest vessel and the furthest part of it 122 µm, rising to 255 once vessels clump. The permeability series is a mouse measurement. The breast distances are not a measurement at all — they are what this project’s own vessel model produces from a density counted in human tumour, so they inherit that density’s standing and add an arrangement nobody has measured. The page says which is which wherever either appears.
Past the plasma membrane the two best-measured stages in the whole path — that roughly 70% of what a cell internalises is sent back out, and that 1–2% of the remainder reaches the cytosol — were measured on lipid nanoparticles carrying siRNA, and are still not lent to a conjugate. One of the two now has a conjugate figure of its own: up to 45% of internalised anti-EGFR antibody returns to the cell surface, fitted in two cell lines, an upper bound rather than a typical value and not human tissue. The other has dissolved rather than been answered. That 1–2% is endosomal escape, the route a lipid nanoparticle takes; a conjugate's payload is freed in the lysosome and leaves by a route the payload itself decides — a cleavable linker releasing membrane-permeable MMAE lets it diffuse out, a non-cleavable maytansinoid needs the lysosomal transporter SLC46A3, and T–DM1's catabolite keeps a charged lysine that stops it crossing at all. So there is no single escape fraction for a conjugate to be missing. The question was the wrong shape.
So the atlas can rank two macromolecules by how much of each reaches a cell surface in a tumour whose drainage has failed, state the assumptions the ranking rests on, and name the measurement that would overturn it. It cannot predict potency, response rate, or anything about an individual patient.
The three measurements this page used to ask for have all been gone after, and it is worth saying what happened to them, because two of the three did not end the way a wanted measurement is supposed to. The pore geometry is confirmed absent in human tissue everywhere, and the published animal figures turn out not to be borrowable even in principle: the same cell line grown at two sites in one mouse differs more than threefold, so the quantity belongs to the host site rather than to the tumour and there is no figure for a tumour type to lend. The conjugate escape fraction dissolved — it is not one quantity, because the route out of the lysosome is set by the linker and the payload's charge. Only the carrier permeability existed, and finding it showed the drop it was meant to refine was never the wall's property to begin with.
What would change what this page can say, now, in order of value. Calibrated receptor copies per cell counted from the side a drug arrives on, which does not exist for any target in this atlas and which is why the number of target molecules drawn in the contact view is a drawing decision in twelve environments out of thirteen. A cytosolic-delivery fraction for a conjugate of a stated linker and payload, together with a recycling fraction in human tissue rather than the two-cell-line fit now held. Whether the perfused vessels carry the radial gradient that total vessel counts show from a tumour's edge to its middle — one measurement, and it decides the shape of the vessel field, which moves the poorly-reached fraction several fold at the same degree of unevenness. And a permeability for a rigid particle, since the carrier series above is flexible polymers and a liposome of the same mass is a different object.
Every number is from published literature, read for its numbers rather than its abstract; nothing in this atlas was measured by us and no molecule described here has been synthesized. Microvessel densities are not comparable between environments as published — some are hotspot counts, some whole-slide random sampling — so each is labeled with its method and they are never placed on a shared axis. Drawings are to scale within a panel, along the dimension the scale bar marks; where a dimension is unpublished the drawing uses a neutral placeholder and the "Not measured" panel says so.