Dual-conditioning campaign for pretrained 3D molecular diffusion

This dataset contains the canonical generation outputs and analysis resultsfor a dual-conditioning study of a pretrained pocket-conditionedthree-dimensional molecular diffusion model. The task combined hardinpainting of a seven-heavy-atom substructure from ligand A withinference-time guidance toward the global shape of ligand B. The campaign comprised three SARS-CoV-2 Mpro ligand pairs, five guidancestrengths, and five random seeds, giving 75 runs and 691 pair-specificevaluations. Deduplication of shared unguided baselines yielded 649 uniquegenerated structures. Of the 648 structures that could be chemicallysanitized, 555 were fragmented. The record includes the generated SDF files, reference ligands, campaignmanifests, frozen and fixed-order directional shape evaluations,unique-structure connectivity and fragmentation tables, condition-levelsummaries, restrained-minimization outputs, figure source data, finalmanuscript figures, environment information, file inventories, andchecksums. No guided structure satisfied the connected strict-dual criterion undereither Shape Protrude convention.