ELF-NET Data S10: Boltz-2 and AlphaFold 3 structure predictions for 3,600 novel dark proteome domains

Complementary Boltz-2 v2.2.1 and AlphaFold 3 v3.0.1 structure predictions for the same 3,600 rank-binned representatives from the Pfam-dark proteome analyzed by ELF-NET. The representatives span ten equal bins of environmental coupling and derive from 33,950 recurrent domain families clustered from 201 million proteins without Pfam assignments in the 2,044-sample mixed-source domain-analysis set. The complete resource contains Boltz-2 PDB structures and confidence files; AlphaFold 3 top-ranked and diffusion-sample CIF structures, per-residue pLDDT, PAE, pTM, fraction-disordered and ranking metrics; sequence and MSA inputs; mapping and rank-bin manifests; summary tables; scripts; and provenance documentation. Both prediction resources cover all 3,600 targets. Related to Figure S8, the dark-proteome structural analysis, and STAR Methods in Coupling of oceanographic state to the dark proteome: a foundation for genome-informed marine productivity modeling.