AMAN-1: data, structures and code for the orthogonal in-silico verification of a de novo PD-L1 pro-binder

Data, structural models, score tables, pre-registered acceptance criteria and analysis codebehind the paper "Orthogonal in-silico verification of a de novo PD-L1 pro-binder: aframework, and the design claims it refuted". Nineteen quantitative design claims about AMAN-1 — an 86-residue de novo binder against thehuman PD-L1 IgV domain, deployed as a de-immunized, protease-activated bivalent pro-binder —were submitted to an orthogonal verification framework resting on four requirements: everypredictor is given a positive and a negative control on the target in question; quantitativeclaims are tested by methods that differ in kind rather than only in implementation;acceptance criteria are fixed before the calculation runs; and both sides of aprotein-protein interface are scored. Seven claims were refuted, two were qualified outright, and ten survived, five of those tenwith a qualification stated in Table 1 of the paper. Every sequence used in the work is in this archive and printed in the paper, including thede-immunized variant V4-DI, which carries three substitutions relative to the published core:S13E, R48T and K58N. Version 1 of this record withheld that sequence; this version does not,and the file that recorded the omission has been removed. The archive contains: the binder, mask, PD-L1, PD-1 and PD-L2 sequences and the residuenumbering maps; structural models covering the core alone and bound to PD-L1, the maskedmonomer, and the activated and masked bivalent constructs at four linker lengths from threeco-folding models; the four starting poses used for the dynamics and metadynamics runs; 120per-model AlphaFold2-Multimer scores with ipSAE, pDockQ, pDockQ2 and LIS computed by twoindependent implementations at matched cutoffs, together with per-model footprint-overlap andparatope-usage counts; the interface contact atlas and its Voronoi-tessellation cross-check;per-frame and per-replicate molecular dynamics and coarse-grained membrane analysis;worm-like-chain linker effective concentrations; class I and class II binding predictions forthe junction peptides; MMP cleavage-site windows; a result memo for every claim in the paper'sTable 1; both pre-registered acceptance criteria verbatim with their outcomes, both of whichfailed; and the analysis scripts, the four notebooks that ran the heavy jobs, the testharnesses that checked their logic before any GPU time was spent, and the vendored referenceipSAE implementation. ALL RESULTS ARE COMPUTATIONAL. No protein was expressed, and no affinity, cleavage rate,expression yield or immunological response was measured. None of the quantities depositedhere is a measurement. One quantity reported in the paper cannot be recomputed from this archive, and the paper saysso: the per-frame contact series of the rigid-linker trajectories was not retained, so neitherof the two irreconcilable aggregates derived from it can be re-derived. Neither is reported.