Features in a nutshell

122 words | 1 min read

One of the pillars of machine learning is extracting useful features for models to learn with. Machine learning features for drug discovery in particular are numerous and rapidly evolving. New techniques are constantly emerging but using tools from different sources quickly becomes tedious, inconvenient, and prone to incompatibilities.

The features submodule aims to address these issues. It is a collection of many different types of featurisers which transform molecules (or chemical structures in general) into a spectrum of machine learning features. Whether you are looking for engineered features (such as fingerprints) or learned features (from neural network embeddings), features provides a standard and modular interface for using these featurisers and also allows you to add your own!