An automated software methodology for biomedical statistics, data pre-processing, and machine learning

Biomedical data generated from clinical informatics, biomarker discovery, and laboratory medicine has substantially increased in the past several years due to improvements in automation, access to software tools, and computational power. Consequently, comprehensive analysis of these data via human evaluation for predictive analytics, early disease detection and diagnosis, personalized medicine, and treatment planning has become increasingly challenging. Application of data pre-processing, statistical analysis, survival analysis, and nuanced machine learning methods requires substantial training and expertise to arrive at meaningful conclusions. The objectives of this study are to develop and demonstrate a software methodology that automates key processes in biomedical data analysis, including preprocessing, statistical evaluation, survival analysis, and machine learning, while requiring no coding expertise.

A novel architecture was designed as a modular wrapper around various open-source R software packages to provide comprehensive data analysis, including data pre-processing, statistical methods, machine learning, and stacked ensemble machine learning.

Three use-case scenarios were explored to illustrate major capabilities: clinical survival analysis, biomarker discovery, and diagnostic model development and validation. The methodology offers user-friendly access to powerful data analytical functions regardless of the user's programming experience.

We anticipate that further testing of the proposed software methodology in clinical and research settings will enable adoption for automated biomedical data analysis.