Visually guided preprocessing of bioanalytical laboratory data using an interactive R notebook (pguIMP).

Area of Science:

• Bioinformatics
• Computational Biology
• Pharmacology

Background:

• Machine learning evaluation of pharmacological data necessitates high data quality.
• Current data preprocessing tools often require programming expertise, limiting accessibility.
• Interactive graphical user interfaces are needed for efficient data cleaning and transformation.

Purpose of the Study:

• To introduce pguIMP, a novel R-based graphical software package for interactive and reproducible bioanalytical data preprocessing.
• To provide a user-friendly tool for non-data analysis experts in biomedical research.
• To enhance the quality and structure preservation of bioanalytical datasets.

Main Methods:

• Development of pguIMP as an R package with a fixed sequence of preprocessing steps.
• Integration of machine learning-based imputation techniques, including k-nearest neighbors.
• Utilizing a Shiny-based web interface for interactive use.
• Evaluation using lipidomics and drug research datasets with k-means and DBSCAN clustering.

Main Results:

• pguIMP enables reproducible, interactive data preprocessing without programming knowledge.
• Machine learning imputation methods in pguIMP preserve data structures like clusters better than classical methods.
• The software successfully processed bioanalytical datasets from lipidomics and drug research.
• The Shiny interface ensures ease of use for researchers without data analysis expertise.

Conclusions:

• pguIMP serves as a valuable, accessible tool for preprocessing bioanalytical data in biomedical research.
• The package facilitates the generation of high-quality, cleaned datasets suitable for machine learning applications.
• pguIMP streamlines the data preprocessing pipeline, enhancing reproducibility and data integrity.