Creating Multiple Sequences Alignments with deep reinforcement learning

Overview

Multiple Sequence Alignment (MSA) refers to the process of aligning biological sequences (DNA, RNA or proteins) for comparison. This is generally done for evolutionary analysis and, more recently, drug design and discovery. Advanced protein predictor networks such as AlphaFold use MSA to predict the 3D structure of proteins. However, as an NP complete problem, MSA is a difficult. This project proposes an innovative method to create MSAs leveraging deep reinforcement learning using a novel operator layer for pairwise feature extraction.

Features

• Single-agent solution: Create an MSA using a single RL agent.
• Multi-agent solution: Create an MSA using two co-operative RL agents.
• Ensemble model: Up to five agents with majority voting.

Installation

1. Clone the repository:

   git clone https://github.com/jovanneste/DeepRL-MSA.git
   cd DeepRL-MSA

2. Install dependencies:

   pip install -r requirements.txt

Usage

Naviagte to the src directory and run a solution.

Single-agent

Run the single-agent solution:

python main_msa.py 

Multi-agent

Run the multi-agent solution:

python main_msa.py --multi

Ensemble

Run the voting solution:

python main_msa.py --vote

Contact

• Author: Joachim Vanneste
• Email: joachimvanneste1@gmail.com