Study findings

The two KEAP1 constructs each have 15 models: seeds 7, 42 and 101, with five samples per seed. The same exact 118-residue binder sequence is used in both arms. All 30 original coordinate files and 30 matching confidence summaries are available below.

Across the 15 numerically matched pairs, the median extended-minus-short iPTM difference is −0.01. Common-core residue-pair contact counts change by medians of +5 at 4 Å and +14 at 5 Å. Contacts are target–binder residue pairs with at least one heavy-atom separation within the stated distance, not atom-pair totals.

After fitting the shared target core, median binder-placement RMSD is 9.959 Å. Seed medians are 3.701 Å for seed 7, 18.027 Å for seed 42 and 6.855 Å for seed 101. Median common-target core-fit RMSD is 0.212 Å and binder self-fit RMSD is 0.420 Å. This combination indicates heterogeneous relative placement despite a much smaller within-chain structural difference.

Neither the added N-terminal segment nor the added C-terminal segment directly contacts the binder in any of the 15 extended-construct models at 4 or 5 Å. Absence of these contacts does not make the two modeling contexts equivalent, because the additional target sequence changes the represented molecular context.

Interpretation and reproducibility

The 30 outputs are dependent computational realizations, not 30 independent experiments. Numerical seed/sample pairing does not isolate every source of stochastic variation.

Contacts, PAE, iPTM and geometry establish neither binding, affinity, NRF2 competition, intracellular activity, NRF2 function nor an SMA-related effect.

Compact inputs preserve exact sequence, construct and seed identities, but omit enriched inference features, complete confidence matrices, multiple-sequence alignments and templates. They cannot reproduce the full inference by themselves.

The older Binder10246 model series is a separate comparison and is not included in these counts.

Dataset download

All 30 models