QSARmil documentation

QSARmil is a Python package for molecular multi-instance machine learning (MIL) applied to QSAR (quantitative structure-activity relationship) modeling.

Instead of representing a molecule as a single feature vector, QSARmil treats it as a bag of instances (e.g. conformers, fragments) and learns directly from that structure — letting the model decide which instances matter most for a given property.

QSARmil offers two levels of access:

• Beginner — a predefined, zero-configuration pipeline. Hand it SMILES and target values, get predictions back. See Installation to get set up, then Tutorials for the full walkthrough.

• Professional — build a custom pipeline from QSARmil’s individual modules (conformer generation, descriptor calculation, MIL methods, hyperparameter optimization, consensus modeling) directly. Each of these is documented in its own page below.

Companion packages milearn (MIL algorithms) and QSARcons (consensus modeling) power parts of the pipeline under the hood — see Ecosystem for how the pieces fit together.

Contents

• Installation
• Training data
• Conformer generation
• Descriptor calculation
• Multi-instance learning
• Hyperparameter optimization
• Consensus modelling
• Key instance detection
• Ecosystem
• Interfaces (CLI)
• Tutorials