Deepmol: an automated machine and deep learning framework for computational chemistry.

Area of Science:

• Computational Chemistry
• Machine Learning
• Cheminformatics

Background:

• Machine Learning (ML) is transforming computational chemistry, but researchers face challenges in algorithm selection, data preprocessing, and feature engineering.
• Automating these ML pipeline steps is crucial for advancing molecular property and activity prediction.

Purpose of the Study:

• To introduce DeepMol, an Automated Machine Learning (AutoML) tool specifically designed for computational chemistry.
• To automate the critical steps of the ML pipeline, including data representation, preprocessing, and model configuration.

Main Methods:

• DeepMol automatically identifies optimal data representations, preprocessing techniques, and model configurations for molecular prediction tasks.
• The tool was benchmarked on 22 datasets, comparing its automated pipelines against traditional, labor-intensive methods.

Main Results:

• DeepMol achieved competitive pipelines on benchmark datasets, rivaling those developed through extensive manual feature engineering and model selection.
• The tool demonstrated superior performance and efficiency across various molecular machine learning tasks.

Conclusions:

• DeepMol offers a groundbreaking, state-of-the-art AutoML solution for computational chemistry, enhancing accessibility and reproducibility.
• Its open-source nature, comprehensive documentation, and support for diverse ML models make it a flexible and robust tool for researchers.