Why brute-force peptide search fails.
Framing the discovery problem before generating any candidate. The combinatorial sequence universe is unworkable without constraints.
Drug design fails when mechanism is asserted, not proven — and combinations are found by trial and error.
Two unsolved problems compound. Search spaces are astronomically large — even short peptides yield ~10¹³ candidates no sampler can cover — and there is no predictive theory of which combinations are stable or super-additive, so polypharmacology stays empirical. DiscoveryLab attacks both: it constrains the search and it machine-checks whether a cross-modal combination is a stable controller.
DiscoveryLab treats drug design as a search-space and mechanism-verification problem, not a text-generation problem. Generation, ranking, rejection, pathway reachability, intervention blockade, and perturbation-evidence checks — and, for combinations, a machine-checked stability proof — happen before a candidate moves toward wet-lab validation. The peptide search-space figures below are our calibrated reference modality.