Summary of the article:

Although interest in using machine learning (ML) to support drug development is growing, the technical barriers associated with complex algorithms have limited widespread acceptance. In response, we developed Pharm-AutoML, an open source Python package that enables users to automate the construction of ML models and predict clinical outcomes, particularly in the field of drug interventions. In particular, our approach simplifies tedious steps in the ML workflow, including data preprocessing, model tuning, model selection, analysis of results, and model interpretation. In addition, our open source package helps identify the most predictable machine learning pipeline among the defined search spaces by selecting the best data preprocessing strategy and setting the ML model meta-parameters. This package now supports multi-class classification tasks and additional functions are being developed. Using a set of five publicly available biomedical datasets, we demonstrate that our Pharm-AutoML performs better than other machine learning frameworks, including H2O with default settings, by showing improved classification prediction accuracy. We further demonstrate how model interpretation methods can be used to help explain the ML pipeline accurately to end users. Pharm-AutoML provides advanced clinical predictions and scientific insights to novice and expert users. What are the important points of studying current knowledge on the subject?

Machine learning (ML) is a powerful way to analyze complex healthcare data, and Auto ML (AutoML) offers a way to automate tedious steps in the ML pipeline. However, current AutoML frameworks do not meet the needs of biopharmaceutical or health researchers due to the lack of end-to-end automation of ML pipelines, including data preprocessing and model interpretation. What question did this study address?

Pharm-AutoML is designed to provide an integrated AutoML solution, enabling biopharmaceutical researchers to automate the construction of ML models, predict clinical outcomes, and interpret ML models. What does this study add to our knowledge?

Pharm-AutoML automates the ML pipeline from data preprocessing, model tuning, model selection, analysis of results, and model interpretation. We show that such a global workflow can perform better than the current implementation of ML and AutoML.

How might this change discovery, development, and / or drug treatment?

Pharm-AutoML helps accelerate the development, deployment, and interpretation of the ML model, thus facilitating the use of ML to predict results and gain insights into clinical trials and pharmaceutical research.