Computational Chemistry and Computational Chemistry

Molecular Representations

SMILES

Concept: Text representation of molecules.

Advantages: Compact, human-readable.

Disadvantages: Not unique, not structured.

Molecular Graphs

Concept: Atoms as nodes, bonds as edges.

Advantages: Structured, well-suited for GNNs.

Applications: GNNs for molecules.

ML Methods

Graph Neural Networks

Models: GCN, GAT, MPNN for molecules.

Applications: Property prediction, classification.

Advantages: Effective for molecules.

Generative Models

VAE: Variational Autoencoders for molecules.

GAN: Generative Adversarial Networks.

Applications: De novo drug design.

Practical Examples

Example 1: Molecular Property Prediction with GNN

Graph: Create a molecular graph from SMILES.

GNN: Train an MPNN for property prediction.

Prediction: Predict properties of new molecules.

Validation: Compare with experimental data.

Example 2: De Novo Drug Design

Generator: Train a VAE or RL for generation.

Generation: Generate new molecules.

Filtering: Filter by properties, validity.

Optimization: Optimize for target properties.