Drug Discovery (Molecular Binding Affinity Prediction)

Project Overview

This project implements a Graph Neural Network (GNN) to predict molecular binding affinities, a crucial task in drug discovery. The solution combines cheminformatics (RDKit) with deep learning (PyTorch Geometric) to analyze molecular graphs.

Key Features

Methodology

Molecular Graph

Project workflow diagram

Architecture

GNN(
  (conv1): GCNConv(5, 64)
  (conv2): GCNConv(64, 64)
  (fc): Linear(in_features=64, out_features=1)
)

Results

Model Performance

📉 MSE: 0.0848
🏋️ Training Epochs: 1000
💻 Hidden Channels: 64

Hyperparameter Tuning

Best Parameters:
- Hidden Channels: 128
- Learning Rate: 0.001
Achieved MSE: 0.0762

Feature importance plot

Conclusion/Future Work

The GNN model demonstrates promising results in predicting binding affinities, achieving an MSE of 0.0848.

Future work includes:

About