Molecular Design

Research notes on the computational molecular design pipeline: generating novel molecules, predicting their properties, and predicting synthesis routes.

Generation – De novo molecular generation (autoregressive, RL-tuned, target-aware, latent-space, search-based) plus benchmark suites and evaluation metrics
Property Prediction – QSAR modeling, molecular property prediction methods, and property prediction benchmarks (MoleculeNet, Activity Cliffs)
Reaction Prediction – Forward synthesis and retrosynthesis prediction

Molecular Generation

Generative models for de novo molecular design, organized by approach: autoregressive generation, RL-tuned optimization, target-aware design, latent-space methods, search-based methods, and evaluation.

Reaction Prediction

Sequence-to-sequence models for forward reaction prediction and retrosynthesis using SMILES-based chemical language modeling.

Property Prediction

Molecular property prediction methods, benchmarks, and evaluation: QSAR models, transfer learning, scoring methods, and benchmark suites like MoleculeNet.