Drug Discovery Applications of Electronic Landscape Stability Diagnostics in Metal-Centered Biological Active Sites

Electronic Landscape Stability Diagnostics for Metal-Centered Biological Active Sites (Version 2.0 adds 7 new locked systems — see README.md for changelog) This is Version 2.0 of the dataset originally deposited at https://doi.org/10.5281/zenodo.19142883. The manuscript submitted to JCTC (ct-2026-00566z) cites the original DOI. Version 2.0 (March 22, 2026) extends the dataset to 16 locked canonical results across Fe, Zn, and Cu metal branches. New results include the Cu⁺ d¹⁰ oxidation-state control (Cu_SOD_minimal_CuI), the full-scaffold Cu²⁺ water-binding result (Cu_SOD_minimal_water), a four-warhead Zn²⁺ CA2 inhibitor chemotype ladder (sulfonamide, hydroxamate, carboxylate, phosphonate), and the first locked metabolic-disease benchmark (Cu_AOC3_minimal, SSAO/VAP-1). Together these results demonstrate that electronic landscape ruggedness is governed by at least three separable contributions: oxidation-state/electron-count class, local coordination geometry, and metal identity. This dataset supports the manuscript: Brahmbhatt, A. "Electronic Landscape Ruggedness as a Reproducibility Diagnostic for Open-Shell Quantum Chemistry: A Cu/Zn Metalloenzyme Benchmark." Journal of Chemical Theory and Computation, submitted March 22, 2026 (Manuscript ID: ct-2026-00566z). Benchmark Classification and Binding-State Perturbation Mapping Using Penalized Ensemble VQE Quantum Clarity LLC March 2026 Abstract We report the application of the Electronic Landscape Stability Diagnostic (ELSD) platform to a series of biologically relevant metal-centered active-site fragments, including Fe(II) porphyrin, Zn²⁺ carbonic anhydrase models, and Cu²⁺ superoxide dismutase (SOD) variants. Using a GPU-accelerated penalized variational quantum eigensolver (VQE) with multi-seed ensemble sweeps, explicit N-sector and Sz-sector enforcement, and full energy decomposition per run, we classify the electronic landscape of each system into one of four regimes: Rigid Stability, Coherent Open-Shell, Multi-Basin, or Model Pathology. Nine locked results are reported across Fe, Zn, and Cu systems, including canonical benchmarks, matched perturbation states, and one diagnostic scaffold control. A four-perturbation controlled series on the Cu²⁺ SOD active site — varying donor protonation state, donor identity, coordination number, and bound ligand — demonstrates that the dominant electronic family is preserved across all 25 runs while ensemble ruggedness spans a 21× range across the series. A matched apo/bound pair shows that water binding increases landscape ruggedness by 9.2× without changing the electronic family, providing the first binding-state classification result in the present Prometheus biological series. These findings establish three platform capabilities: (1) identification of stable, decision-grade biological benchmarks across metal centers and spin states; (2) causal mapping of perturbation-dependent ruggedness within a preserved electronic family; and (3) discrimination between fragments sharing the same electronic family but representing materially different screening models. Plain Language Summary What we did We built a computational tool that examines the electronic structure of metal-centered active sites in biological molecules — the kind found at the heart of enzymes involved in drug metabolism, cancer biology, and cellular defense. Instead of running a single simulation and accepting the result, we ran dozens of independent calculations on the same target, each starting from a different random point, and asked: do they all agree? We tested this across three different metal centers — iron (Fe), zinc (Zn), and copper (Cu) — in models that represent real drug-discovery targets. We then systematically changed one thing at a time in the copper system: the chemical character of a nearby donor molecule, the number of coordinating atoms, and whether a water molecule was bound to the metal center. Across nine locked results and twenty-five Cu perturbation runs, we recorded what happened to the electronic landscape each time. What we discovered Three things that we did not expect to find together: 1. The same tool works across very different chemical systems. Iron, zinc, and copper active sites — with different electron counts, different spin states, and different coordination environments — can all be classified by the same methodology. The electronic landscape of a closed-shell zinc site and an open-shell copper site look very different, but the tool correctly identifies and distinguishes them without needing to be reconfigured for each one. 2. Ruggedness is tunable, and the baseline is not always the most stable point. When we changed the chemical character of the donor molecule attached to the copper center — making it weaker, stronger, or removing one donor entirely — the electronic landscape changed in a systematic but non-obvious way. The unmodified baseline turned out to sit at a local maximum of ruggedness. Every perturbation we tested, in any direction, produced a tighter, more predictable landscape. This means the most "natural" configuration is not necessarily the most stable one electronically — a finding with real implications for how drug targets are modeled. 3. A bound ligand changes the electronic model even when it does not change the electronic family. When we added a single water molecule to the most stable copper configuration, the dominant electronic state remained identical — the same quantum mechanical solution appeared in every run. But the spread of solutions across runs increased by 9×. In practical terms: the apo (unbound) site and the water-bound site look the same qualitatively, but they are not the same model for screening purposes. One is nine times more variable than the other. That difference is invisible to conventional single-point methods. Why this matters Drug discovery programs that involve metal-centered targets — which includes a large fraction of enzyme inhibitor programs — typically assume that the computational model of the target is settled before screening begins. Our results show that this assumption deserves scrutiny. The electronic character of a metal active site can change substantially depending on protonation state, coordination number, and bound ligand, even when the changes look small on paper. More specifically: two target models can produce the same dominant electronic solution and still represent materially different screening landscapes. A team that treats the apo and ligand-bound states as equivalent is optimizing compounds against a target that may be 9× more electronically variable than their reference model suggests. The practical implication is a new kind of quality-control step that can be inserted before large-scale screening: classify the electronic landscape of the target model first, confirm it is stable and coherent, and only then commit to the screening campaign. This does not replace existing computational methods — it sits upstream of them, on the question of whether the model itself is trustworthy. 1. Platform Description 1.1 Engine All calculations were performed using the Prometheus VQE engine (prometheus_vqe_engine_penalized.py), a GPU-accelerated penalized variational quantum eigensolver built on TorchQuantum and OpenFermion-PySCF. Key parameters: Parameter Value Ansatz UCCSD depth 6 Active space 10 electrons / 10 orbitals (20 qubits) Basis set LANL2DZ (effective core potential) Pauli terms ~3100 per system Hardware NVIDIA L40S GPU (44 GB) Convergence conv=1e-6 (default), conv=1e-7 (tighter sweeps) Max iterations 300 (default), patience=30 1.2 Sector enforcement Number-sector enforcement: penalty term λ_N × (N̂ − N_target)² added to Hamiltonian. Spin-sector enforcement: penalty term λ_Sz × (Ŝz − Sz_target)² added to Hamiltonian. Standard operating points: λ_N = 2.0 Ha (Zn/Cu systems), λ_Sz = 5.0 Ha (open-shell Cu). 1.3 Quality criteria A result is locked only when all five criteria are met simultaneously across the full seed ensemble: Sector clean: |⟨N⟩ − N_target| < 0.1 and |⟨Sz⟩ − Sz_target| < 0.1 Dominant determinant probability: dom_p > 0.99 Same dominant bitstring family across all seeds Penalty contribution audited and confirmed numerical noise only (decompose_energy.py) σ and range consistent with claimed landscape classification 1.4 Classification regimes Regime σ (kcal/mol) Description Rigid Stability typically < 0.5 Single tight basin, decision-grade Coherent Open-Shell typically 0.5–2.0 Broader but single-family, trustworthy Multi-Basin > 10 Multiple competing families, sector-clean Model Pathology N/A Sector escape despite enforcement 2. Locked Results — Nine Total Three categories: canonical benchmark systems, locked perturbation states, and locked diagnostic controls. 2.1 Full results registry Canonical benchmark systems (6): System Metal σ (kcal/mol) Seeds Regime Status FePorphyrin_FeII ls Fe²⁺ d⁶ low-spin 0.3375 5 Rigid Stability LOCKED ✓ Zn_CA2 minimal Zn²⁺ d¹⁰ tetrahedral 0.0933 9 Rigid Stability LOCKED ✓ Zn_CA2 imidazole Zn²⁺ d¹⁰ + 3 imidazole 0.4367 5 Rigid Stability LOCKED ✓ Cu_SOD_minimal Cu²⁺ d⁹ 3N+1O 1.7035 5 Coherent Open-Shell LOCKED ✓ Cu_SOD_2imidazole Cu²⁺ d⁹ 2N+1O 0.0803 5 Rigid Stability LOCKED ✓ Cu_SOD_2imidazole_water Cu²⁺ d⁹ + axial H₂O 0.7419 5 Coherent Open-Shell LOCKED ✓ Locked perturbation states (2): System Metal σ (kcal/mol) Seeds Regime Status Cu_SOD_protonated Cu²⁺ d⁹, donor neutralized 0.6402 5 Tightened open-shell LOCKED ✓ Cu_SOD_acetate Cu²⁺ d⁹, donor strengthened 0.9017 5 Tightened open-shell LOCKED ✓ Locked diagnostic control (1): System Metal σ (kcal/mol) Seeds Regime Status Zn_squareplanar Zn²⁺ d¹⁰ D4h artificial 43.29 5 Multi-Basin (scaffold) LOCKED DIAGNOSTIC 2.2 FePorphyrin_FeII low-spin Fragment: Fe²⁺ + 4 porphyrin N (square planar, Fe-N=2.01 Å) + axial H₂O. Config: charge=0, mult=1 (singl