H3ACE Workshop Notebooks

This repository contains Jupyter notebooks for machine learning tasks on molecular property prediction.

Session 1

SS_Decision_Trees.ipynb

Predicts aqueous solubility (LogS) of organic compounds for drug discovery.

• Dataset: ESOL (1,128 compounds with SMILES and solubility values)
• Model: Decision Tree Regressor on RDKit descriptors
• Task: Regression
• Predicted: LogS (log solubility in mol/L)

SS_MLP.ipynb

Classifies organic compounds as soluble or poorly soluble.

• Dataset: ESOL (SMILES and LogS values)
• Model: Multi-Layer Perceptron (PyTorch)
• Task: Binary Classification
• Predicted: Solubility Class (0/1)

Session 2

mpnn.ipynb

Predicts molecular properties using Message Passing Neural Networks.

• Dataset: ESOL
• Model: Simple Mpnn
• Task: Regression
• Predicted: LogS and LogP

gcn.ipynb

Predicts molecular solubility and lipophilicity with Graph Convolutional Networks.

• Dataset: ESOL
• Model: GCN with convolution and pooling layers
• Task: Regression
• Predicted: LogS and LogP

gat.ipynb

Predicts physicochemical properties using Graph Attention Networks.

• Dataset: ESOL
• Model: GAT on molecular graphs
• Task: Regression (multi-target)
• Predicted: LogS and LogP