Enhancing Reinforcement Learning in 3-Dimensional Hydrophobic-Polar Protein Folding Model with Attention-Based Layers

This repository contains the implementation and experimental results for the paper "Enhancing Reinforcement Learning in 3-Dimensional Hydrophobic-Polar Protein Folding Model with Attention-Based Layers". The study explores the use of attention-based mechanisms to improve reinforcement learning strategies for solving the HP protein folding problem in 3D.

---

Repository Structure

• Attn_DQN_3d_hp.py
  This file contains the implementation of the attention-enhanced Deep Q-Network (DQN) used in the experiments.
  Note: Parameters, including the protein sequence, need to be adjusted directly within the script as there is currently no command-line interface.

• Best_Runs/
  This folder contains:

  • Training Logs: Detailed logs for all training sessions.
  • Best Results: Optimal solutions. These results include all the best solutions (with identical energies but varying structures) in the best_result/ subfolder. Coordinates are saved in best_result/best_results_log.csv

---

Sample Best Results

Here are some examples of the best results obtained during the experiments:

Length: 20, Energy = -11

Length: 24, Energy = -13

Length: 60, Energy = -49

Length: 60, Energy = -49 (Alternative Structure)

---

Install the dependencies using:

  pip install -r requirements.txt