A Quantum Framework for Protein Binding-Site Structure Prediction on Utility-Level Quantum Processors

TL;DR

The work tackles local protein binding-site structure prediction for short, flexible fragments, a regime where classical and DL methods struggle. It introduces an end-to-end quantum framework that maps sequences to a tetrahedral backbone lattice and encodes constraints into a sparse Pauli Hamiltonian, optimized via Variational Quantum Eigensolver (VQE) with an EfficientSU2 ansatz. A two-stage execution strategy mitigates noise by separating energy estimation from measurement decoding, enabling hardware-executable predictions on utility-level quantum processors. Hardware experiments on IBM’s 127-qubit processor show quantum predictions with lower RMSD and stronger docking affinities than AlphaFold3 and classical baselines, highlighting practical viability and a path toward quantum-assisted docking and design workflows. The results validate a concrete, hardware-feasible quantum pipeline for structural biology with potential impact on drug discovery and protein engineering.

Abstract

Accurate prediction of protein active-site structures remains a central challenge in structural biology, particularly for short and flexible peptide fragments where conventional and simulation-based methods often fail. Here, we present a quantum computing framework specifically developed for utility-level quantum processors to address this problem. Starting from an amino acid sequence, we formulate structure prediction as a ground-state energy minimization problem using the Variational Quantum Eigensolver (VQE). Amino acid connectivity is encoded on a tetrahedral lattice model, and structural constraints-including steric, geometric, and chirality terms-are mapped into a problem-specific Hamiltonian represented as sparse Pauli operators. Optimization is performed with a two-stage architecture that separates energy estimation from measurement decoding, enabling noise mitigation under realistic device conditions. We evaluate the framework on 23 randomly selected protein fragments from the PDBbind dataset and 7 fragments from therapeutically relevant proteins, and execute experiments on the IBM-Cleveland Clinic quantum processor. Predictions are benchmarked against AlphaFold 3 (AF3) and classical simulation-based approaches using identical postprocessing and docking procedures. Our method outperforms both AF3 and classical baselines in RMSD (root-mean-square deviation) and docking efficacy. These results demonstrate an end-to-end, hardware-executable pipeline for biologically relevant structure prediction on real quantum processors, highlighting its engineering feasibility and practical advantages over existing classical and deep learning approaches.

Figures (14)

• Figure 1: End-to-End Hardware-Executable Quantum Workflow for Protein Structure Prediction. The framework takes an amino acid sequence as input, encodes structural constraints into Pauli operators, and constructs a parameterized quantum circuit using a selected ansatz. A hybrid VQE is executed on IBM Quantum hardware via Qiskit Runtime to optimize circuit parameters. The lowest-energy binary outcome is reverse-mapped to a predicted backbone structure, followed by postprocessing for downstream applications. The framework consists of five submodules: (1) Problem Modeling Module: Models the input sequence as Pauli operators and constructs the corresponding Hamiltonian. (2) Quantum Circuit Construction Module: Selects the circuit architecture, builds a parameterized quantum circuit, and associates it with the Hamiltonian. (3) Hybrid Variational Optimization Module: Optimizes the parameters of the quantum circuit using a quantum-classical hybrid algorithm (VQE) to search for low-energy states. (4) Measurement Module: Reconstructs a fixed-parameter quantum circuit using the optimized parameters, performs measurements, and decodes the results. (5) Postprocessing Module: Processes the predicted outputs to generate complete protein structures suitable for downstream biological tasks such as docking or simulation.
• Figure 2: Details of two example cases in real-world application scenarios. (A–D) Case1, 6mu3 (YAGYS): (A) The quantum circuit composed of native gates of the quantum processor, generated after compiling the modeled problem for Case 1. (B) The distribution plot of the measured quantum states, where the tallest bar represents the most frequently observed outcome. This highest-probability result corresponds to the final expected measurement outcome. (C) The actual qubits occupied by Case 1 on the quantum processor. (D) VQE energy trace with five lowest-energy conformers (insets: docking affinity, RLB, RUB). (E–G)Case2, 2xxx (GAVEDGATMTFF): (E) The quantum circuit composed of native gates of the quantum processor, generated after compiling the modeled problem for Case 2. (F) The actual qubits occupied by Case 2 on the quantum processor. (G) VQE energy trace with six lowest-energy conformers (insets: docking metrics).
• Figure 3: Structural testing and visual representation (A) Docking workflow. Predicted structures from both methods are hydrogenated, charge-neutralized, and docked against the same receptor using an identical postprocessing pipeline. (B) Native structure of 3b26. (C) Docking pocket view. (D) Quantum overlay. The quantum prediction accurately follows the fragment trajectory and backbone conformation, showing high consistency with the reference structure. (E) AF3 overlay. The AF3 prediction tends to converge to a local minimum under limited sequence information. (F) RMSD comparison. Structural deviation between predicted and reference structures; smaller RMSD indicates higher accuracy. (G) Affinity comparison. (H) RMSD upper bound. (I) RMSD lower bound.
• Figure 4: Scalability test of the prediction framework. Sliding-window quantum prediction of A$\beta$42. (A) Segmentation and two-stage VQE workflow (7-residue window, one-residue stride) Repeatedly invoking the framework to predict multiple fragments and then integrating them into a complete protein structure. (B) Reassembled full-length A$\beta$42 model shown in surface, cartoon, and ball-and-stick views. (C) Backbone vector density map; high-density regions mark the most probable conformations. Lighter-colored regions indicate lower variance across multiple predictions, suggesting that repeated invocations did not introduce significant deviations in the final structure. The overall prediction error remains within an acceptable range.
• Figure S1: RMSD comparison between quantum and AF3 predictions. In 18 out of 23 test cases, quantum-predicted structures exhibit lower RMSD values relative to experimental structures compared to AF3, indicating improved geometric fidelity.