Working laboratory
A medicine is a particular arrangement of atoms, and altering part of that arrangement changes the molecule in two quite different ways at once. Some things follow immediately and exactly: what the new molecule is made of, what it weighs, how many places on it can donate or accept a hydrogen bond. Other things do not follow at all. Whether it still binds its target, whether it reaches the tissue that target sits in, and whether it is safe are questions that take experiments, and no experiment has been run on a molecule that does not yet exist.
This page calculates the first kind and declines the second, and keeps them in separate panels so the difference stays visible while you work. There are two ways in. You can change the chemical structure of a real medicine, or you can leave the structure alone and change an assumption about one stage of how the drug does its work.
A medicine is not designed in the abstract. It is designed to reach something, somewhere, and the tissue it has to cross decides a great deal before any question of what the molecule is made of.
A target is a protein, and the single most important thing about it is where it sits. A protein on the outside of a cell, a protein facing a lumen behind a sheet of cells, and a protein in the cytoplasm are three completely different design problems, and only one of them can be solved with the kind of molecule this lab builds.
Thirteen with a verified structure. Each is a working answer to a real problem, and the useful question is what happens when you change part of one.
Drag one onto any dashed position on the molecule. The R marks the bond that would join it. Adding a group replaces a hydrogen, which is the only edit that cannot leave an atom with the wrong number of bonds.
Every medicine the atlas holds a record for in this tissue, ordered by how far it got. A journey is a series of gates between the blood and the target; these are how many each one cleared, and which gate stopped it. Open any of them to read the whole journey. This is what happened, not a prediction about what you are building.
Suggestions drawn from the tissue and target you chose and from the molecule in front of you. Each says why, and each can be applied in one click so you can see what it does and undo it.
Three answers, kept apart on purpose, because running them together is how a drawing starts sounding like a prediction. What the change has settled, what these groups are generally observed to do, and what nobody has measured.
Every structure on this page is a connectivity table checked on each build against its molecular formula and the valence of each of its atoms, by frontend/scripts/check-lab.mjs. That script also applies all twelve groups at every available position on every medicine, some two thousand substitutions in all, and refuses to pass if a single result fails to balance. The charge each medicine carries in blood is checked against the charge it is known to carry. Nothing in this mode computes a potency, a permeability or an outcome in a tissue, because this project holds no measurement that would support one.
For a chemotherapy to kill a cancer cell, five things have to happen in order. Enough of the dose has to arrive where the cell is. Once there it has to become, or remain, the active form. The active form has to engage whatever it acts on. That engagement has to produce real damage rather than merely binding. And enough of that damage has to survive the cell’s own efforts to repair and absorb it. A medicine that is excellent at four of those and poor at the fifth does nothing at all, which is the whole reason for thinking in chains.
This model carries three of those chains, because three are what the published evidence supported building. Choose the one whose kind of damage you are interested in. It is not a choice of medicine; it fixes which causal story the model will reason about.
Each slider is one stage of that chain. Every one starts at 1.0, which means the same chosen coefficient as the parent model. These controls ask how sensitive the model is to a coefficient; they do not measure how good a molecule is. Moving one above 1.0 states an assumption: that some future molecule would do better at that stage. The useful way to work is one stage at a time, because the question worth asking is which single stage has to improve before the model’s answer changes.
This names the set of assumptions above so you can save it and come back to it. It is not a molecule, and nothing from the structural side is carried into it: the two modes answer different questions and are kept separate on purpose.
The chain below is relative to the parent. Green means you have assumed the candidate does better at that stage than the medicine it came from, and amber means worse. Remember that the last stage runs the other way: the cell buffering less damage is the outcome that favours your candidate.