QA-PBP-RL-Nexus

This repository contains code and data to generate and validate plastic binding peptides (PBPs) for the manuscript: "Developing Peptide-Based Strategies for Microplastic Pollution via a Nexus of Biophysical Modeling, Quantum Computing, and Artificial Intelligence."

Overview

RL_PPO_Peptides/ contains code to discover and generate peptides in solution space.

QA_peptide_generator/ contains code to generate optimized amino acid sequences for unique backbones.

demo/ contains full-body code and sample data with sample inputs that help you generate peptides via both QA and PPO.

Peptide design QA + RL/ folder contains the following data to generate and validate different peptides and results claimed in the manuscript. The organization of the folder is explained below:

• Peptide Design QA+RL/AllDesigns.xlsx: contains the best designs for each system conformation for all plastics using either quantum annealing (QA), PepBD, or Proximal Policy Optimization (PPO). There is a separate tab for the plastics polyethylene (PE), polypropylene (PP), polystyrene (PS), and polyethylene terephthalate (PET). For each system conformation, the best score and the corresponding amino acid sequence are provided. Note that for polyethylene, QA and PPO do not have solutions for all conformations.

• Peptide Design QA+RL/AminoAcidFrequencies.xlsx: contains the amino acid frequencies for QA designs for all four plastics and PepBD designs for polyethylene.

• Peptide Design QA+RL/Example_PPO_Score_Trajectories:provides two examples of the evolution of the peptide score over the course of PPO.

• Peptide Design QA+RL/MD_Data.xlsx: provides the adsorption enthalpy (dH) and adsorption free energy (dG) for peptides, with values calculated using the MM/GBSA method. Separate tabs are provided for PE, PP, PS, and PET. Each tab lists results for peptides found by QA, PepBD, PPO, or generation of a random amino acid sequence. Each entry lists the peptide amino acid sequence, the adsorption enthalpy, and adsorption-free energy.

• Peptide Design QA+RL/Peptide_Properties.csv: contains the Camsol solubility score and the net peptide charge for peptides designed by QA, PepBD, and PPO. For QA and PPO designs, the system conformation corresponding to the design is listed.

• Peptide Design QA+RL/PottsModel_Energies: contains all one-body energies (SingleEnergy.txt) and two-body energies (Pairwise.txt) for all system conformations (or "bb"s) for PE, PP, PS, and PET.

• Peptide Design QA+RL/PPO_All_Sequences_All_Conformations: contains all solutions found by PPO for all sampled system conformations. The results for each conformation are in a separate folder named "bb". As described in the manuscript, a sampled amino acid sequence is only considered a solution if its corresponding score is within 5 of the best scores found by QA.

• Peptide Design QA+RL/PPO_NumUnique_vs_Score.csv: contains the total number of solutions found per system conformation, as well as the best score found by QA for that conformation.

• PPO_Seq_2_SideChainEnvironment_Analysis: contains analysis of the relationship between side chain geometric environment (SideChainEnvironment.csv) and the most frequent amino acid found in the environment (SequenceAnalysis.csv).

• Peptide Design QA+RL/PPO_Seq_2_SideChainEnvironment_Analysis/SequenceAnalysis.csv: for each system conformation, lists the best score found by PPO, the number of solutions found, and the most common amino acid at each of the 12 residues in the peptide.

• Peptide Design QA+RL/PPO_Seq_2_SideChainEnvironment_Analysis/SideChainEnvironment.csv: for each system conformation, provides the distance between the beta carbon and the top of the surface (COM), the angle between the alpha carbon-beta carbon vector and the surface normal vector (Angle), and the solvent-accessible surface area (SASA) for each of the 12 residues.

System Requirements

Operating System

This repository has been developed and tested on macOS, ensuring full compatibility with this operating system. However, the provided scripts are designed to be platform-independent and should work seamlessly on other versions of macOS and Windows.

Hardware Requirements

The PPO models do not require any non-standard hardware and can be run on a typical computer. The code also provides the option to utilize a GPU via the PyTorch library for speedup in the reinforcement learning process. Quantum Annealing is performed on the D-Wave quantum computer. An API token from D-Wave is required to successfully run the QA part of the code.

The code has been tested on a system with the following specifications:

• Apple M1 Pro
• 16GB of RAM
• macOS Sonoma

Package Requirements

• python 3.9.7
• pandas 2.0.0
• pytorch 2.0.0
• numpy 1.24.2
• dimod 0.10.15
• dwave-hybrid 0.6.10
• tqdm 4.64.1
• openpyxl 3.1.2

Installation Guide

The source code is not distributed as a package, therefore, no installation is required.

Clone the repository: Clone the repository to a local directory using the following command:

https

git clone https://github.com/PEESEgroup/QA-PBP-RL-Nexus.git

Demo

Detailed examples of how to use our model to generate PBPs using Proximal Policy Optimization (PPO) are provided in demo/PPO_pbp.py; This Python file contains the base code and basic hyperparameters required to train the policy for peptide generation. Additionally, demo/ppo_pbp_generation.py contains the code to generate peptides using the learned policy. GPU support using CUDA is provided to enhance performance, and we have attempted compatibility with PyTorch for efficient GPU utilization.

Detailed examples of how to use our model to generate PBPs using Quantum Annealing are provided in demo/qa_pbp_generation.py. This Python file contains the base code required to generate new peptides. Access to D-Wave hardware via an API token is necessary to utilize the Quantum Annealer. For demo purposes, we have used a basic sampler to replicate the workflow.

Instructions for Use

To reproduce the results presented in our paper, please follow these steps:

1. Navigate to the demo/ directory and examine the readme section and provided .py files with sample inputs.
2. Ensure that your environment meets the package and system specifications outlined in the instructions provided within the repository.

Note: Due to the stochastic nature of the computational models, you may not generate identical peptides in multiple runs. However, you can expect to observe the same distributions of peptide properties (such as total energy and amino acid distribution) as reported in the paper, provided the same design parameters are used.

Citation

@article{,
author = {Jeet Dhoriyani, Michael T. Bergman, Carol K. Hall, and Fengqi You},
title = {Developing peptide-based strategies for microplastic pollution via a nexus of biophysical modeling, quantum computing, and artificial intelligence},
journal = {},
volume = {},
number = {},
pages = {},
doi = {},
abstract = {}}