It can tell you the molecule scored well. For a filing, a patent, or a synthesis decision, the provenance of a candidate matters as much as its score, and a learned model does not have one to give.
a learned generator produces a structure without a derivation. There is nothing to audit, reproduce, or defend later. Design rationale, geometric pipeline · our position
stochastic sampling gives a different answer on a different seed, so the same request does not yield the same candidate set. Design rationale · our position
from plausibility. A structure that scores well can still be chemically invalid, and the score does not check. Chemical validity checker · our position
over an explicit energy gives a sampler whose distribution is stated rather than learned, so the acceptance criterion is inspectable. Property-targeted graph sampler · our position
This is the same argument as everywhere else on this platform. The answer matters, and so does whether you can show how it was reached. A geometric pipeline is auditable by construction; a learned one is not, whatever its benchmark says.
Six that close the biggest gaps. 5 in the full molecular and drug discovery library.
The gap it closes. Explicit-energy detailed-balance MCMC, so the sampling distribution is a stated property rather than an emergent one.
See it run in your browser →The gap it closes. Valence, charge and connectivity checked deterministically. A generator that never sees this can propose structures that cannot exist.
See it run in your browser →The gap it closes. Resolves structure geometrically rather than sampling toward it, so the same input yields the same output every time.
See it run in your browser →The gap it closes. A distance you can define is a distance you can defend in a similarity argument. A learned embedding gives a number without a metric you chose.
See it run in your browser →The gap it closes. Screening under an explicit acquisition rule, so the reason a candidate was selected next is recorded rather than inferred.
See it run in your browser →These are computational chemistry tools, not a discovery program and not a substitute for synthesis, assay or a medicinal chemist. Validity checking establishes that a structure is chemically well-formed, which is a much weaker claim than that it is synthesisable, stable, or active.
Every tool above runs entirely in this browser tab. Nothing is uploaded, nothing is sent to a model, and each result cites the rule it applied.
Free tier, no card, no install. An account saves your work and lets you export a citable record of a run. See all 5 molecular and drug discovery tools.