Microbiome-based correction for random errors in nutrient profiles derived from self-reported dietary assessments.

Area of Science:

• Nutritional Epidemiology
• Computational Biology
• Bioinformatics

Background:

• Dietary intake assessment in large cohorts relies on self-reported data, prone to significant measurement errors.
• Inaccuracies in nutrient profile calculations arise from these errors, limiting epidemiological study reliability.
• Existing computational methods to correct these dietary assessment errors are scarce.

Purpose of the Study:

• To introduce a novel deep-learning approach, Microbiome-based Nutrient profile corrector (METRIC), for correcting random errors in self-reported dietary assessments.
• To evaluate METRIC's performance in minimizing errors for nutrient profiles, especially those influenced by gut microbial metabolism.
• To assess the impact of gut microbial composition on METRIC's error correction capabilities.

Main Methods:

• Developed a deep-learning model (METRIC) integrating gut microbial data to correct dietary assessment errors.
• Utilized 24-hour recalls and diet records as input for dietary assessment correction.
• Validated METRIC using synthetic datasets and three real-world datasets.

Main Results:

• METRIC demonstrated excellent performance in minimizing simulated random errors in nutrient profiles.
• The model showed particular effectiveness for nutrients metabolized by gut bacteria.
• METRIC maintained significant error correction capabilities even when gut microbial composition data was excluded.

Conclusions:

• METRIC offers a promising computational solution for enhancing the accuracy of self-reported dietary assessments in nutritional epidemiology.
• The approach shows potential for improving nutrient profile calculations, even without direct microbiome data.
• Further investigation is needed to confirm METRIC's efficacy in correcting real-world measurement errors in dietary assessment instruments.