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

Since dietary intake is challenging to directly measure in large-scale cohort studies, we often rely on self-reported instruments (e.g., food frequency questionnaires, 24-hour recalls, and diet records) developed in nutritional epidemiology. Those self-reported instruments are prone to measurement errors, which can lead to inaccuracies in the calculation of nutrient profiles. Currently, few computational methods exist to address this problem. In the present study, we introduce a deep-learning approach— M icrobiom e -based nu t rient p r of i le c orrector (METRIC), which leverages gut microbial compositions to correct random errors in self-reported dietary assessments using 24-hour recalls or diet records. We demonstrate the excellent performance of METRIC in minimizing the simulated random errors, particularly for nutrients metabolized by gut bacteria in both synthetic and three real-world datasets. Additionally, we find that METRIC can still correct the random errors well even without including gut microbial compositions. Further research is warranted to examine the utility of METRIC to correct actual measurement errors in self-reported dietary assessment instruments. Here, the authors introduce METRIC, a deep-learning method that corrects measurement errors in self-reported dietary nutrient profiles using assessed data and microbial composition, effectively enhancing dietary accuracy.