Transforming Big Data into AI-ready data for nutrition and obesity research

Area of Science:

• Obesity and Nutrition Research
• Data Science
• Bioinformatics

Background:

• Big Data offers significant potential for advancing obesity and nutrition research.
• Raw Big Data requires extensive preprocessing for use in machine learning (ML) and artificial intelligence (AI) models.
• Preprocessing is the most complex stage, demanding ML, human judgment, and specialized software.

Purpose of the Study:

• To detail the preprocessing pipelines for popular obesity/nutrition Big Data sources.
• To highlight challenges and decision impacts in creating AI- and ML-ready data.
• To emphasize the critical need for end-user understanding of data preprocessing.

Main Methods:

• Review of three major obesity/nutrition Big Data sources: microbiome, metabolomics, and accelerometry.
• Detailed examination of preprocessing pipelines and associated specialized software.
• Analysis of how preprocessing decisions influence the final AI- and ML-ready data products.

Main Results:

• Identified specific preprocessing steps, software, and challenges for microbiome, metabolomics, and accelerometry data.
• Demonstrated the impact of preprocessing choices on the quality and usability of AI- and ML-ready data.
• Presented opportunities for improving quality control, preprocessing speed, and data consumption.

Conclusions:

• Big Data holds promise for discovering novel modifiable factors in obesity research.
• Transparency in the AI- and ML-ready data preparation process is essential for accurate interpretation.
• Clear understanding of preprocessing complexities is vital for investigators and clinicians.