qsarmil 1.0.3

Molecular multi-instance machine learning

QSARmil is a package for designing pipelines for building QSAR models with multi-instance machine learning algorithms.

Introduction

Multi-instance machine learning for molecules

Installation

pip install qsarmil

Benchmarking QSARmil

To facilitate benchmarking QSARmil against alternative platforms, we developed a meta-model builder that eliminates the need for manual adjustments to the model-building protocol. The pipeline automatically generates multiple multi-conformer models using diverse descriptor sets and multi-instance learning methods, and then applies a genetic algorithm to identify the optimal consensus combination of individual models. The input data should be provided as a pandas DataFrame, where the first column contains the molecular SMILES strings and the second column contains the corresponding target property values.

from qsarmil.meta import MultiConformerModel

model = MultiConformerModel(num_conf=10, hopt=True, task="regression", output_folder="mcf", verbose=True)
y_pred = model.run_predict(df_train, df_test)

Use cases

See the examples of QSARmil application for different tasks in the tutorial collection .