Project README

Overview

This project is designed for processing, filtering, and optimizing molecular SMILES strings using machine learning models. It includes tools for SMILES cleaning, latent space encoding/decoding, and optimization.

Prerequisites

• Python 3.8+
• requirements.txt

Installation

1. Install dependencies:

   pip install -r requirements.txt
   cd hgraph2graph
   pip install .
   cd ..

2. Unzip this file to the root directory and rename it to "data". Link - https://drive.google.com/file/d/1R24W4YYKVeVc6G-EdGMnFRFMJWJOeo9e/view?usp=sharing. It contains csv file with antioxidants and their pseudo labling.

1. Regression model

Used from our RL algorith to pseudo label smiles. Example of usage in get_smiles.ipynb

2. Optimisation algorithm

file - gt_converted.py

Variables inside:

path_to_vocab - path to vocabulaty of the model VAE

path_to_data_to_train_MLP - path to csv file with pseudo labeled smiles

File Descriptions

• data_preparation.py: Cleans and filters SMILES strings based on chemical properties.
• gt_converted.py: Encodes SMILES into latent vectors, trains an MLP, and optimizes latent vectors.
• get_smiles.ipynb: Predicts target values for SMILES and extracts top candidates.

Example Workflow

1. Run gt_converted.py
2. Follow the steps in get_smiles.ipynb

Notes

• Ensure rdkit is properly installed for SMILES processing.
• Adjust hyperparameters in gt_converted.py for better optimization results.
• Original repo with VAE - https://github.com/wengong-jin/hgraph2graph/tree/master