Answers are grounded in this paper and cite their sources.
De novo peptide binders are attractive leads for biological pesticides because they can be designed against validated insecticide targets without a small-molecule discovery campaign. Two generative pipelines dominate this space: BindCraft2, an AlphaFold2/ProteinMPNN campaign that accepts designs only after a cascade of structural filters, and RFD4-Proteina, a flow-matching foundation model that generates conditioned backbones and sequences in a single pass. We benchmarked both on three insecticide targets: acetylcholinesterase (PDB 1QON, 540 residues), the ecdysone receptor ligand-binding domain (1R1K, 236 residues) and the ryanodine receptor SPRY2 domain (6KIM, 169 residues). Every design was re-folded with the same AlphaFold2 oracle. RFD4 returned 300 designs in 3.3 GPU-hours, 82 of which passed a common re-folding screen. BindCraft2 returned 10 designs in 58.7 GPU-hours, none on two of the three targets, and 3 of its 10 passed the same screen. The advantage of RFD4 was entirely concentrated on the smallest, contiguous target: there it delivered 82 usable designs in 0.32 GPU-hours, against 3 in 1.95 GPU-hours from BindCraft2. On the two larger targets RFD4 complexes looked bound in the generated coordinates but did not survive re-folding, and BindCraft2 accepted nothing. Generation speed and re-foldable interface quality therefore separated on this target set: RFD4 is the more cost-effective generator, but neither method produced validated binders on two of the three targets.
No takes yet. Share an insight, caveat, or question.
Fu et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: