Pharm-AutoML: An open-source, end-to-end automated machine learning package for clinical outcome prediction

Area of Science:

• Pharmacology
• Computer Science
• Bioinformatics

Background:

• Machine learning (ML) adoption in drug development is hindered by complex algorithms.
• Automating ML workflows is crucial for wider application in pharmacological research.

Purpose of the Study:

• To introduce Pharm-AutoML, an open-source Python package for automating ML model construction in drug development.
• To streamline and improve the prediction of clinical outcomes from pharmacological interventions.

Main Methods:

• Pharm-AutoML automates data preprocessing, model tuning, selection, analysis, and interpretation.
• It identifies optimal ML pipelines by optimizing preprocessing and hyperparameters.
• The package currently supports multiclass classification tasks.

Main Results:

• Pharm-AutoML demonstrated superior predictive accuracy compared to other ML frameworks like H2O on five biomedical datasets.
• Model interpretation methods were successfully applied to explain the ML pipeline's decisions.
• The package enhances clinical predictions and provides scientific insights.

Conclusions:

• Pharm-AutoML simplifies ML for drug development, benefiting both novice and expert users.
• The automated approach improves predictive accuracy and interpretability of ML models.
• This tool facilitates wider adoption of ML in pharmacological research and clinical outcome prediction.