Research code for geometric obstruction in cyclic peptide backbones.
-
Reference-connection strain — the residual SE(3) screw of the product of residue transforms with every residue at preferred Ramachandran geometry — is the geometrically right obstruction object (
derive.py). It is strongly sequence-dependent (poly-Pro ≫ poly-Ala). -
Bishop holonomy of the deposited CA curve is a real, reproducible geometric invariant (transport ↔ Gauss–Bonnet to 1e−15) but is not the sheaf θ of the abelian paper. On AF-designed closed scaffolds it shows approximately zero Spearman correlation with reference strain (ρ ≈ 0, n=450) — descriptive association, not a claim of statistical independence under a pre-registered null.
-
ψ ≈ φ closure locus — pure geometry: 182 uniform rise≈0 solutions for N=5–16, 100% with |ψ−φ| < 15°.
-
Combinatorial proline count is not a free topological invariant.
-
Rosetta total score does not track backbone strain; P2 against it is invalid.
See STATUS_P1_P2.md, CORRECTIONS.md, CORRECTION_NOTE.md, and NOTES_EMPIRICAL.md.
Operating call (post-merge): use E_ref as a sequence prior — do not force P2 on deposited AF structures. See NOTES_EREF_PRIOR.md.
pip install numpy scipy
python derive.py # reference strain on poly-A / sequence variants
python middle_out_v2.py # slice 0 = E_ref (not n_Pro)
python run_p1.py # Bishop on experimental CCDC CIFs (needs network)
python scripts/eref_prior.py # sequence prior: exp vs hall vs random (needs data/*.csv)| Test | Result |
|---|---|
| E_ref vs Bishop E (n=450) | ρ ≈ 0 |
| E_ref vs Rosetta total | ρ ≈ −0.10 |
| Empirical rama NLL/residue vs N (n=131) | ρ ≈ −0.62 (length association; see NOTES_EMPIRICAL) |
| E_ref vs NLL/residue | ρ ≈ +0.36 (descriptive; density fit on same set) |
P2 is parked until a backbone-visible energy (rama/omega terms, cyclized−linear ΔE) or literature yields exist. Do not chase deposited AF totals. Details: P2_PROTOCOL.md, NOTES_EREF_PRIOR.md.
derive.py— reference-connection strain (omega_mode='trans'|'cis_pro')holonomy_extract.py/run_p1.py— Bishop measurementmiddle_out_pipeline.py— SE(3) kinematic slicesSTATUS_P1_P2.md/CORRECTIONS.md— status and corrections logNOTES_EMPIRICAL.md/NOTES_CIS_PRO.md/NOTES_TIER_A_LITE.md— empirical notesNOTES_EREF_PRIOR.md— E_ref as sequence prior (designed vs exp vs random); P2 parkedP2_PROTOCOL.md— pre-registered P2 (deferred until backbone-visible energy or yields)CODE_REVIEW.md— multi-domain review
This project is licensed under the PolyForm Noncommercial License 1.0.0. Commercial use requires a separate license from the copyright holder, B. Jones (b.jones@jtech.ai). See NOTICE.