Editorial disclosure

AI assisted primary-source inspection, evidence synthesis and drafting. The user selected the slate and authorized publication. This preparation does not claim completed human source verification; supplements, raw data and code were not independently audited or reproduced.

What the paper claims

The authors report that generated thiolation domains support product formation across multiple NRPS architectures and that selected designs improve yield relative to native-domain controls.

What was tested

Pretrained ESM3, ProteinMPNN and EvoDiff models were integrated with iterative design–build–test–learn cycles and data-guided prioritization. Recombinant variants were tested in minimal, full-length and hybrid NRPS systems using product-formation assays including LC–MS, with biochemical characterization and molecular-dynamics simulations.

What the evidence supports

ESM3, ProteinMPNN and EvoDiff were coupled to iterative design–build–test–learn cycles: 76 designed thiolation domains and 578 recombinant NRPS variants tested across minimal, full-length and hybrid systems. Product-formation assays, including LC–MS, found selected designs with up to approximately threefold higher yields than native-domain controls. Biochemical characterization of one representative design and molecular-dynamics simulations complement the functional experiments.

Limitations

Compatibility depends strongly on the neighboring domain: success does not transfer uniformly across architectures. The threefold result is a best-case improvement, not an average across designs; some screens use relative LC–MS peak areas. Simulation-derived mechanisms are hypothesis-generating rather than causally established. No clinical efficacy is tested.

Open-access abstract and main-text results/discussion read; publisher data/code statements checked. Supplementary/source files are listed but not audited. Custom code is author-reported on GitHub and Zenodo; raw LC–MS and simulation trajectories are available only on request under the stated terms. Code was not executed and independent replication was not assessed. Evidence scope: AI inspection of the abstract and selected full-text sections; no independent replication, exhaustive supplementary review or completed human source verification is claimed.

Sources

Corrections & versioning

Current version 1. Initial evidence brief with explicit source scope, limitations and missing information.

No later corrections logged.