Protein Structure\nPrediction with High Degrees of\nFreedom in a Gate-Based Quantum Computer

Protein folding, which traces the\nprotein three-dimensional (3D)\nstructure from its amino acid sequence, is a half-a-century-old problem\nin biology. The function of the protein correlates with its structure,\nemphasizing the need to study protein folding to understand the cellular\nand molecular processes better. While recent AI-based methods have\nshown significant success in protein structure prediction, their accuracy\ndiminishes with proteins of low sequence similarity. Classical simulations\nface challenges in generating extensive conformational samplings.\nIn this work, we develop a novel turn-based encoding algorithm with\nmore significant degrees of freedom that successfully runs on a gate-based\nquantum computer and predicts the structure of proteins of varied\nlengths utilizing up to 114 qubits (IBM hardware). To make the problem\ntractable in quantum computers, the protein sequences were described\nwith the simplistic HP model (H = hydrophobic residues, P = polar\nresidues). The proposed formulation successfully captures the so-called\nnucleation step in protein folding, the hydrophobic collapse, that\nbrings the hydrophobic residues to the core of the protein.