New multi-agent framework uses quantum-classical loops to improve protein structure prediction

Researchers propose QFoldAgent, a closed-loop multi-agent framework for 5-residue tetrahedral-lattice protein folding that addresses a key limitation in hybrid quantum-classical protein structure prediction: manual tuning of Hamiltonian penalty weights.

The system consists of three agents: a design agent that proposes sequence-conditioned penalty weights, a VQE-based quantum-classical pipeline that optimizes the Hamiltonian under Qiskit Aer noise, and a feedback agent that uses energy-landscape diagnostics and MolProbity validation signals to refine penalties across cycles.

Ground-truth metrics such as RMSD are withheld from the agents and used only for evaluation, ensuring that improvements reflect genuine optimization behavior rather than overfitting to known structures.

On a benchmark of 55 QDockBank-derived fragments with known structures, QFoldAgent reduces median RMSD from 3.64 Å to 3.20 Å, with the largest gains observed on the hardest targets.

On 100 coverage-optimized unseen sequences, the closed loop raises structural validity from 87.5% to 98.7%, recovers 87% of initially invalid cases, and the strongest controller improves cycle-3 energy on 87% of sequences while maintaining 96% Ramachandran-favored geometry.

The authors argue that iterative agent control can systematically improve optimization behavior and reduce failure cases in short-sequence quantum folding settings.