QFoldAgent: An Autonomous Quantum Optimization Multi-Agent System for Protein Structure Prediction

Computer Science > Artificial Intelligence

Submitted on 11 May 2026

Authors: Winson Chen, Yuqi Zhang, Sixu Chen, Nuo Xu, Qiang Guan, Caiwen Ding

Abstract

Hybrid quantum-classical approaches to protein structure prediction remain highly sensitive to Hamiltonian penalty weights. However, existing lattice-based workflows typically fix these coefficients manually and evaluate only very short fragments in simulation. We present QFoldAgent, a closed-loop multi-agent framework designed for 5-residue tetrahedral-lattice folding. In this system, a design agent proposes sequence-conditioned penalties, a VQE-based quantum-classical pipeline optimizes the resulting Hamiltonian under Qiskit Aer noise, and a feedback agent leverages energy-landscape diagnostics alongside MolProbity validation signals to refine penalties across iterative cycles. Ground-truth metrics—such as RMSD—are never exposed to the agents and are used solely for evaluation.

We evaluate the framework on two complementary datasets: 55 QDockBank-derived fragments with known structures and 100 coverage-optimized unseen sequences. On the QDockBank benchmark, QFoldAgent reduces median RMSD from 3.64 Å to 3.20 Å, with the most substantial improvements observed on the hardest targets. For unseen sequences, the closed-loop system raises structural validity from 87.5% to 98.7%, recovering 87% of initially invalid cases. Additionally, the strongest controller improves cycle-3 energy on 87% of sequences while maintaining 96% Ramachandran-favored geometry. These results demonstrate that iterative agent control can systematically enhance optimization behavior and reduce failure cases within a 5-residue quantum setting.

Subjects: Artificial Intelligence (cs.AI)

Cite as: arXiv:2607.22549 [cs.AI]

DOI: https://doi.org/10.48550/arXiv.2607.22549

via ArXiv AI