Analogical generation · prediction · active learning
We design new molecules from data you already have — and rank which to run next.
We mine analogies from your library, build molecules from them, predict what they do, and rank them for your next experiments. The results rejoin the library, and the next pass reasons from more.
Two things, or two changes — the transfer is the same move.
An analogy is a transfer: what your library has already resolved, carried onto a question it has not. The rules are mined as analogies, the molecules built by applying them, the predictions carried across.
Stage one · rules
We mine the rules out of your own library, including ones nobody has tried yet.
Each one arrives with the evidence behind it. Every rule is mined algorithmically from your own data, never carried in from a generic set.
- Demonstrated. A change your compounds have already made, with the evidence behind it and how consistently it held.
- Implied. A change your data supports but nobody has made — related options your compounds have each used, never against each other.
Stage two · generation
Those rules then build new candidates from molecules you choose.
Every candidate carries the analogy that produced it.
The rule behind it, the molecule it started from, the evidence for both, and whether the edit was checked by chemistry or carried by precedent alone.
A chemist can disagree with a candidate for a reason, and name it. The loop only compounds if each turn can be audited, so the reasoning travels with the molecule.
Stage three · prediction
Each candidate is scored against the measured compounds its rule came from.
Ranking needs numbers, and ours are carried from precedent: what a measured case already tells you, applied to the change in question. The same analogy that built a candidate is the one that scores it.
That is also what makes the hard cases tractable. A smooth model interpolates across an activity cliff and flattens it; carrying a measured change across transfers the jump from a case that already crossed it.
Stage four · acquisition and return
You run the candidates worth testing, and their results become the next round of precedent.
They are ranked for acquisition, not only for score: which experiments buy the most for the next turn.
The library is read again, over data it did not have before. New rules appear, and implied changes become demonstrated, carrying a number for the first time.
A one-shot model is as good on its last day as on its first. A loop is not.
Every experiment deepens a library of analogies nobody else holds, and that is what makes the next pass better.