Small-molecule optimization
Proposes alternatives to part of a seed compound — a candidate that could bind where the original does, not a bigger version of it — using CReM over a real ChEMBL fragment database. Use once a lead already exists and the question is "what else could work here."
Small-molecule optimization
What it does
Takes one seed compound and proposes alternatives by replacing part of it — answering "what else could bind here," rather than Small-molecule generation's "how could this be extended." Each candidate is a real molecule with one region genuinely different from the seed, not just a minor variant.
Each candidate comes back with QED (drug-likeness), SA score (synthetic accessibility), and similarity to the seed (0 to 1, where 1 would mean identical) — that last number is specific to this tool, since "how different is this from where I started" is exactly the question being asked here.
How candidates are generated
Same underlying fragment-context matching Small-molecule generation uses, applied to an existing fragment inside the molecule rather than an open position — a region of the seed is swapped for an alternative seen in an identical local chemical environment elsewhere in the reference database.
Works best on a full candidate compound, not a small building block. Confirmed directly during evaluation: the same settings on a very small seed can end up replacing almost the whole molecule, producing candidates only loosely related to where you started. A real lead compound gives this tool enough to work with; a small fragment doesn't. If the seed is small, the run includes a warning saying so.
Reading the output
Similarity to seed is the number worth checking first — a very low value means the candidate diverged more than expected, worth confirming it's still a fit for the original question. QED and SA score work the same way they do for Small-molecule generation.
Input
One seed compound, saved as an SDF.
Important limitations
No pocket-awareness, same as Small-molecule generation — pair with the Small-molecule docking tool downstream for a target-directed answer.
QED and SA score are heuristics, not predictions — see Small-molecule generation's own doc for the same caveat, which applies here identically.
A small seed compound gets a warning, not an automatic adjustment. This tool doesn't change its own settings based on the seed's size; it flags the concern and leaves the decision to the user.