What problem does it solve?
This Skill addresses common challenges and pitfalls in building machine learning models with molecular data, ensuring robust and reliable results.
Core Features & Use Cases
- Scaffold Splitting: Prevents data leakage by ensuring molecules with the same core structure are not present in both training and testing sets.
- Fingerprint Selection Guidance: Recommends appropriate molecular fingerprints (e.g., Morgan, RDKit) based on the specific ML task (similarity search, QSAR, etc.).
- Pitfall Avoidance: Highlights and provides solutions for common errors like ignoring fingerprint types, not handling parse failures, and feature matrix shape issues.
- Use Case: When developing a QSAR model, use this Skill to perform a scaffold-based split and select the most informative Morgan fingerprints to avoid overestimating model performance.
Quick Start
Use the mol-ml skill to split a list of SMILES strings by scaffold, ensuring no data leakage between training and testing sets.