Master Reaction Prediction

A comprehensive machine learning project for predicting chemical reaction outcomes using deep learning and graph neural networks.

Overview

This repository contains code and datasets for predicting the outcomes of chemical reactions using state-of-the-art machine learning techniques. The project leverages multiple deep learning frameworks including PyTorch, TensorFlow, and specialized chemistry libraries to build predictive models.

Tech Stack

• Core ML: PyTorch 2.2.1, TensorFlow 2.15.0, JAX 0.3.25
• Chemistry: RDKit, DeepChem 2.8.0, PyTorch Geometric 2.6.1
• Data Processing: Pandas, NumPy, Scikit-learn
• Utilities: Lightning, WANDB (experiment tracking), Loguru (logging)
• GPU Support: CUDA 11.8

Installation

Using Conda

conda env create -f conda-env.yaml
conda activate master

Project Structure

Master_Reaction_Prediction/
├── src/                      # Source code
├── datasets/                 # Data files
├── bash/                     # Shell scripts for running experiments
├── visualizations/           # Generated plots and results
├── conda-env.yaml            # Conda environment specification
├── requirements.txt          # Python dependencies
├── illegal_smiles.txt        # Validation rules for SMILES strings
└── .gitignore                # Git ignore rules

Usage

TBD - Add specific usage instructions based on your main scripts in the src/ directory.

Data

The datasets/ directory contains the chemical reaction data used for training and evaluation. See illegal_smiles.txt for validation constraints on molecular SMILES strings.

Development

Environment Setup

1. Create the conda environment: conda env create -f conda-env.yaml
2. Activate: conda activate master

Running Experiments

Scripts for running experiments are located in the bash/ directory.

Results & Visualizations

Visualizations and results from model evaluations are saved to the visualizations/ directory.