Molecule solubility predictor

The brief

Poor solubility kills many drug candidates late in development. Predict the water solubility of molecules from their structure using RDKit descriptors and fingerprints, compare models, and show which structural features matter. Good looks like: a scaffold-based split, models compared fairly, and chemically sensible explanations.

What you will build

1. Load molecules and compute RDKit descriptors and fingerprints
2. Split by scaffold so test molecules are new
3. Train linear, random forest and gradient-boosted models
4. Evaluate with RMSE and R²
5. Explain which features drive solubility

Free, open data to build it with

• AqSolDB: aqueous solubility dataset ↗ Scientific Data (Nature)
• MoleculeNet datasets ↗ MoleculeNet

Check each dataset's licence before you publish your work. Your mentor gives your team its own version of the brief (data, scope or region), so every team's project is its own.