DeepMol

DeepMol automates machine learning pipelines for computational chemistry, including data representation identification, pre-processing method selection, and model configuration for molecular property and activity prediction.

Key Features:

• Automated Data Preprocessing: Automatically selects and applies optimal data pre-processing methods for molecular datasets.
• Adaptive Feature Engineering: Rapidly identifies the most effective data representation for specific molecular property prediction problems.
• Algorithm and Model Configuration Selection: Automatically determines optimal algorithms and model configurations without manual intervention.
• Benchmark Validation: Evaluated on 22 benchmark datasets with competitive performance against manually engineered pipelines.

Scientific Applications:

• Molecular Property Prediction: Predicts molecular properties through automated machine learning pipeline optimization.
• Molecular Activity Prediction: Predicts molecular activity endpoints using automated data representation and model selection.

Methodology:

Automated machine learning pipeline involving identification of data representations, pre-processing methods, and model configurations evaluated on benchmark datasets.