ibp-enm 0.1.0

Spectral Elastic Network Model for protein structural analysis — 7-instrument thermodynamic band classification with 100% archetype accuracy on the 12-protein benchmark.

IBP-ENM

Spectral Elastic Network Model for Protein Structural Analysis

IBP-ENM extracts per-residue structural role profiles, unsupervised domain boundaries, hinge sites, and multi-perspective flexibility predictions from a single static protein structure — no training data, no homology templates, no MD simulations.

The Thermodynamic Band extends the core elastic network analyzer with a 7-instrument carving protocol that classifies proteins into structural archetypes (enzyme, barrel, allosteric, dumbbell, globin) achieving 100% accuracy on the 12-protein benchmark corpus.

Key Features

• Zero-parameter domain detection — Fiedler vector sign changes partition the contact graph into structural domains without any tunable thresholds.
• 7 thermodynamic instruments — Each probes a distinct physical signal (symmetry breaking, resonance sensitivity, diffusion-optimal cuts, entropy disruption, cooperativity, allosteric reach, thermal fragility).
• Two synthesis lenses — EnzymeLens (92% accuracy) and HingeLens (100% accuracy) fuse instrument votes via Meta-Fick diffusion consensus.
• Self-contained — Only requires numpy, scipy, and requests. Fetches PDB structures directly from RCSB.

Installation

pip install ibp-enm

Or from source:

git clone https://github.com/Earthform-AI/ibp-enm.git
cd ibp-enm
pip install -e .

Quick Start

Classify a protein

from ibp_enm import run_single_protein

result = run_single_protein("2LZM", chain="A", verbose=True)
print(result["band_identity"])   # → "enzyme_active"
print(result["hinge_identity"])  # → "enzyme_active"

Core structural analysis

from ibp_enm import IBPProteinAnalyzer, fetch_pdb_ca_data

coords, bfactors = fetch_pdb_ca_data("1MBO", "A")
analyzer = IBPProteinAnalyzer()
result = analyzer.analyze(coords, bfactors)

print(f"Domains: {result.n_domains}")
print(f"Hinge residues: {result.hinge_indices}")
print(f"B-factor correlation: {result.consensus_correlation:.3f}")

Run the full 12-protein benchmark

from ibp_enm import PROTEINS, GROUND_TRUTH, run_single_protein

for pdb_id, chain, name in PROTEINS:
    r = run_single_protein(pdb_id, chain, name)
    expected = GROUND_TRUTH.get(pdb_id, "?")
    got = r["hinge_identity"]
    match = "✓" if got == expected else "✗"
    print(f"  {match} {name:20s}  expected={expected:20s}  got={got}")

Architecture

IBPProteinAnalyzer          ← Core ENM: analyze / compare / probe / listen
    │
ThermodynamicBand           ← 7-instrument orchestrator
├─ 7 × ThermoInstrumentCarver
│      algebraic   — max |Δgap|         (symmetry breaking)
│      musical     — max mode_scatter   (resonance sensitivity)
│      fick        — FickBalancer       (diffusion-optimal cut)
│      thermal     — max |Δτ|           (entropy disruption)
│      cooperative — max |Δβ|           (cooperativity probe)
│      propagative — max spatial_r      (allosteric reach)
│      fragile     — max bus_mass       (thermal soft spots)
│
├─ MetaFickBalancer         ← Consensus / disagreement fusion
│   ├─ EnzymeLensSynthesis  ← 92% accuracy (D110)
│   └─ HingeLensSynthesis   ← 100% accuracy (D111)
│
└─ SurgeonsHandbook         ← Initial snapshot diagnosis

The 12-Protein Benchmark

PDB	Protein	Archetype	Band	Enzyme Lens	Hinge Lens
2LZM	T4 Lysozyme	enzyme_active	✓	✓	✓
1MBO	Myoglobin	globin	✓	✓	✓
2DHB	Deoxyhemoglobin	globin	✓	✓	✓
1GGG	Galactose Oxidase	barrel	✓	✓	✓
2POR	Porin	barrel	✓	✓	✓
4AKE	Adenylate Kinase	enzyme_active	✗	✗	✓
1ANF	ABP (open)	allosteric	✓	✓	✓
3CLN	Calmodulin	dumbbell	✓	✓	✓
1LFG	Lactoferrin	allosteric	✗	✓	✓
1OMP	OmpA	barrel	✓	✓	✓
1HNF	Inorganic PPase	enzyme_active	✓	✓	✓
5CYT	Cytochrome c	globin	✓	✓	✓

Accuracy: Band 83% → EnzymeLens 92% → HingeLens 100%

API Reference

Core Analysis

Symbol	Description
IBPProteinAnalyzer	Core ENM analyzer with analyze(), compare(), probe(), listen()
IBPResult	Dataclass holding analysis results (domains, hinges, B-factors, etc.)
fetch_pdb_ca_data(pdb_id, chain)	Fetch Cα coordinates + B-factors from RCSB PDB
search_rcsb(query)	Search RCSB for PDB entries
gnm_predict(coords)	GNM baseline B-factor prediction

Thermodynamic Band

Symbol	Description
ThermodynamicBand	7-instrument orchestrator class
run_single_protein(pdb_id, chain)	Run full band pipeline on one protein
ThermoInstrumentCarver	Individual instrument carver
ThermoReactionProfile	Per-instrument reaction data
MetaFickBalancer	Vote fusion engine
EnzymeLensSynthesis	Enzyme-calibrated lens (92% accuracy)
HingeLensSynthesis	Hinge-calibrated lens (100% accuracy)

Archetypes & Data

Symbol	Description
ProteinArchetype	Enum: enzyme_active, barrel, allosteric, dumbbell, globin
PROTEINS	The 12-protein benchmark corpus
GROUND_TRUTH	Expected archetype for each benchmark protein
SurgeonsHandbook	Initial snapshot diagnosis generator

Thermodynamic Observables

Symbol	Description
vibrational_entropy(evals)	Vibrational entropy from eigenvalue spectrum
heat_capacity(evals, T)	Heat capacity from mode occupation
helmholtz_free_energy(evals, T)	Helmholtz free energy
spectral_entropy_shannon(evals)	Shannon entropy of the eigenvalue distribution
per_residue_entropy_contribution(evecs, evals)	Per-residue entropy decomposition
hinge_occupation_ratio(evecs, hinge_indices)	Mode participation at hinge sites
domain_stiffness_asymmetry(evals_d1, evals_d2)	Stiffness asymmetry between domains

Testing

pip install pytest
pytest tests/ -v

All 50 tests pass on Python 3.10–3.12.

Citation

If you use IBP-ENM in your research, please cite:

@article{byrom2025ibpenm,
  title   = {IBP-ENM: Spectral Elastic Network Fingerprints for
             Protein Conformational Analysis},
  author  = {Joshua Byrom},
  year    = {2025},
  url     = {https://earthform.ai/papers/ibp_enm_spectral_protein.pdf}
}

@article{byrom2025thermoband,
  title   = {Thermodynamic Band Classification of Protein
             Structural Archetypes},
  author  = {Joshua Byrom},
  year    = {2025},
  url     = {https://earthform.ai/papers/thermo_band_archetypes.pdf}
}

License

MIT — see LICENSE.