Quantum Reinforcement Learning in Protein Folding

This paper focuses on developing an innovative solution to the protein folding problem, a long-standing challenge in Bioinformatics, through the lens of quantum reinforcement learning (QRL). It proposes a hybrid method that incorporates a Quantum Neural Network created using a Parameterized Quantum Circuit. The study is conducted on the Hydropolar Model, a simplified representation of proteins in terms of their polarity, and the experiments are carried out in both 2D and 3D lattice representations. The datasets used for analysis include benchmark proteins from existing research and PDB or FASTA files from the Protein Data Bank. The core of the paper presents the detailed process of modeling and implementing the protein folding problem using QRL, along with detailed comparisons with classical deep reinforcement models inspired by the literature. The findings from this study demonstrate that QRL performs effectively in predicting energies close to optimal levels. However, it poses a greater challenge in terms of convergence when compared to classical methods. Besides the novel method proposed, we also present a supporting custom tool - the protein-folding-gym-utils python library - designed specifically to assist researchers in applying reinforcement learning techniques to the Protein Folding Problem.