Pre-registered hypothesis and prediction record: recursive-collapse trace signature of protein structural bistability in tau and α-synuclein

Pre-registered hypothesis and prediction record for blind topological collapse analysis comparing native disease proteins and their aggregation-causing single-residue variants in two well-characterized neurodegenerative-disease protein systems, to be performed by the DON Research engine. Pre-registration locked on 2026-05-10; experiment will execute only after this record receives a CERN/Zenodo timestamp. Substrates pre-specified (variants will be constructed by single-residue substitution on the fetched native sequence at experiment time): - Pair A: tau (UniProt P10636, MAPT 2N4R isoform, 441 aa) vs tau-P301L (Pro→Leu at position 301; frontotemporal dementia variant, well-documented aggregation propensity) - Pair B: α-synuclein (UniProt P37840, 140 aa) vs α-synuclein-A53T (Ala→Thr at position 53; familial Parkinson's variant, well-documented aggregation propensity) Methodology: each substrate is fed as raw single-letter sequence into recursive_collapse(features, backbone=True, audit_log=True, node_history=True) where features = [[ord(c)] for c in seq]. The engine state is locked at the current numpy path; no engine modification, no parameter tuning, no substrate re-selection between this pre-registration and execution. Disease-state labels (native vs aggregation-causing variant) are bookkeeping conventions external to the engine, which only ever receives the sequence as ASCII byte values. Canonical structural objects are extracted by tools/extract_post_collapse_targets.py with default arguments. Reference baseline (already observed in this lab on 2026-05-10, KRAS WT and four single-residue oncogenic variants run through identical pipeline configuration): - KRAS-WT (188 aa): terminates coherently at scale 38, closure_horizon length=1, 3 polarity_flip_zones, 1 silence_corridor, anchor_core members at psi = −tanh(1.0) = −0.7616 and phi_collapse < 1.0. - KRAS-G12D (G→D at position 12, oncogenic, biological consequence: constitutively active GTP-locked KRAS): terminates at scale 42, closure_horizon length=6, 5 polarity_flip_zones, 2 silence_corridors, anchor_core has 4 members with phi_collapse = 75.0 = ord('K') (kernel un-modulated by any coherent peak event). - KRAS-G12V (G→V at position 12, oncogenic, biological consequence: same as G12D): terminates at scale 44, closure_horizon length=6, 7 polarity_flip_zones, 2 silence_corridors, anchor_core has 3 members with phi_collapse = 75.0 = ord('K'). - KRAS-G12C (G→C at position 12, oncogenic but less aggressive, retains some GTPase activity, sotorasib drug target): terminates at scale 39, closure_horizon length=1, 5 polarity_flip_zones, 2 silence_corridors, anchor_core 9 members with mixed psi (−0.63 to −0.93). - KRAS-Q61K (Q→K at position 61, switch II hotspot variant): terminates at scale 39, closure_horizon length=1, 3 polarity_flip_zones, 2 silence_corridors, anchor_core 4 members with psi = 0.0. The KRAS-G12D and KRAS-G12V signatures are interpreted at the trace layer as recurring oscillation between attempted basin commitments — the substrate's recursive collapse trajectory cannot lock its kernel coherently and instead extends past WT's termination scale with multiple late peak_psi_var resurgences (scales 38–41 for G12D; scales 40–42 for G12V) that fail to commit, leaving terminal anchor_core residues with un-modulated input phi values. The biological consequence in the cell (no GTP off-switch, constitutive activation) and the engine's trace-layer observation (kernel cannot lock) are interpreted as the same phenomenon at different observational layers. Headline prediction — aggregation-prone single-residue variants of intrinsically disordered proteins (tau-P301L and α-synuclein-A53T) should produce an oscillation signature on the trace layer analogous to KRAS-G12D/G12V's, distinguishable from their native-sequence counterparts on at least 3 of the following 5 pre-specified metrics: 1. variant n_scales > native n_scales (extended trajectory) 2. variant len(polarity_flip_zones) > native (more regime oscillations) 3. variant len(silence_corridors) > native (recurring loss of coherence) 4. variant closure_horizon.length > native (delayed lock) 5. variant anchor_core has at least one member with phi_collapse equal to ord(member.aa) within ±0.5 (un-modulated kernel residue, signature of failure to coherently commit) Confirmation rule: H1 confirmed if BOTH pair A and pair B independently show variant strictly greater than native on at least 3 of the 5 metrics. Partially confirmed if exactly one pair satisfies the ≥3-of-5 criterion (reportable but requires additional substrate pairs before claiming generality). Falsified if neither pair satisfies, or if variants show MORE coherent terminations than natives. In all three outcomes the full canonical-objects output for all four substrates will be deposited as raw artifacts; no selective reporting. Files to be deposited POST-experiment alongside this pre-registration record: - tau_native_with_history.json — tau MAPT 2N4R native engine output, full audit + node_history - tau_p301l_with_history.json — tau P301L variant engine output, full audit + node_history - asyn_native_with_history.json — α-synuclein native engine output, full audit + node_history - asyn_a53t_with_history.json — α-synuclein A53T variant engine output, full audit + node_history - tau_native_post_collapse_targets.json — canonical structural objects (anchor_core, polarity_flip_zones, silence_corridors, closure_horizon, etc.) for tau native - tau_p301l_post_collapse_targets.json — same for tau P301L - asyn_native_post_collapse_targets.json — same for α-synuclein native - asyn_a53t_post_collapse_targets.json — same for α-synuclein A53T - 2026-05-10_PROTEIN_BISTABILITY_PREREG.md — this pre-registration record, byte-identical to the version timestamped here This deposit externalizes the hypothesis and predictions BEFORE the experiment runs, providing a third-party-verifiable timestamp via Zenodo's CERN-backed archival infrastructure. The pre-registration is locked at this deposit's DOI. Any modification post-registration requires a new pre-registration record explicitly noting and superseding this one. The KRAS reference-baseline observations cited above were generated 2026-05-10 in the same lab session and are deposited separately (see kras_*_with_history.json and kras_*_post_collapse_targets.json artifacts). Limitations: this experiment is N=2 substrate pairs. The trace-layer pattern's statistical generality across the broader class of aggregation diseases (Alzheimer's amyloid-β, ALS TDP-43/SOD1, Huntington's polyQ, prion diseases, type-II diabetes IAPP, etc.) cannot be claimed from these two pairs alone — extension to other systems requires its own pre-registration and its own deposit. The choice of P301L (over P301S, R406W, K280Δ for tau) and A53T (over E46K, H50Q for α-synuclein) is based on literature documentation depth, not on prior trace-layer evidence. The 5 comparison metrics were selected from patterns observed in the KRAS-WT vs KRAS-oncogenic-variant comparison; their generality across protein families is itself part of what the experiment tests. The KRAS reference baseline is N=1 protein system; the interpretation that KRAS-G12D/G12V's failure-to-terminate signature represents the same trace-layer phenomenon as bistability in aggregation diseases is the working hypothesis under test, not a previously-established result.