QFoldAgent: Autonomous Quantum Multi-Agent System for Protein Folding
A new framework called QFoldAgent uses a closed-loop multi-agent system for hybrid quantum-classical protein structure prediction on 5-residue tetrahedral lattices. A design agent proposes sequence-conditioned Hamiltonian penalty weights, a VQE-based pipeline optimizes under Qiskit Aer noise, and a feedback agent refines penalties using energy-landscape diagnostics and MolProbity validation. Ground-truth RMSD is withheld from agents. Tested on 55 QDockBank fragments and 100 unseen sequences, QFoldAgent reduced median RMSD from 3.64 Å on the benchmark.
Key facts
- QFoldAgent is a multi-agent framework for protein structure prediction.
- It uses a design agent, VQE-based quantum-classical pipeline, and feedback agent.
- The system operates on 5-residue tetrahedral lattices.
- Hamiltonian penalty weights are adapted per sequence.
- Optimization uses Qiskit Aer noise model.
- Feedback uses energy-landscape diagnostics and MolProbity.
- RMSD is used only for evaluation, not exposed to agents.
- Tested on 55 QDockBank fragments and 100 unseen sequences.
- Median RMSD reduced from 3.64 Å on QDockBank benchmark.
Entities
Institutions
- arXiv
- Qiskit Aer
- MolProbity
- QDockBank