The Influence of an AI-Driven Personalized Nutrition Program on the Human Gut Microbiome and Its Health Implications

Abstract

Background/Objectives: Personalized nutrition programs enhanced with artificial intelligence (AI)-based tools hold promising potential for the development of healthy and sustainable diets and for disease prevention. This study aimed to explore the impact of an AI-based personalized nutrition program on the gut microbiome of healthy individuals. Methods: An intervention using an AI-based mobile application for personalized nutrition was applied for six weeks. Fecal and blood samples from 29 healthy participants (females 52%, mean age 35 years) were collected at baseline and at six weeks. Gut microbiome through 16s ribosomal RNA (rRNA) amplicon sequencing, anthropometric and biochemical data were analyzed at both timepoints. Dietary assessment was performed using food frequency questionnaires. Results: A significant increase in richness (Chao1, 220.4 ± 58.5 vs. 241.5 ± 60.2, p = 0.024) and diversity (Faith's phylogenetic diversity, 15.5 ± 3.3 vs. 17.3 ± 2.8, p = 0.0001) was found from pre- to post-intervention. Following the intervention, the relative abundance of genera associated with the reduction in cholesterol and heart disease risk (e.g., Eubacterium coprostanoligenes group and Oscillobacter) was significantly increased, while the abundance of inflammation-associated genera (e.g., Eubacterium ruminantium group and Gastranaerophilales) was decreased. Alterations in the abundance of several butyrate-producing genera were also found (e.g., increase in Faecalibacterium, decrease in Bifidobacterium). Further, a decrease in carbohydrate (272.2 ± 97.7 vs. 222.9 ± 80.5, p = 0.003) and protein (113.6 ± 38.8 vs. 98.6 ± 32.4, p = 0.011) intake, as well as a reduction in waist circumference (78.4 ± 12.1 vs. 77.2 ± 11.2, p = 0.023), was also seen. Changes in the abundance of Oscillospiraceae_UCG_002 and Lachnospiraceae_UCG_004 were positively associated with changes in olive oil intake (Rho = 0.57, p = 0.001) and levels of triglycerides (Rho = 0.56, p = 0.001). Conclusions: This study highlights the potential for an AI-based personalized nutrition program to influence the gut microbiome. More research is now needed to establish the use of gut microbiome-informed strategies for personalized nutrition.

Keywords: artificial intelligence; gut microbiome; human health; personalized nutrition.

Figure 1

An overview of the PROTEIN study design and host parameters. (A) A schematic overview of the study design. Twenty-nine healthy participants used the personalized nutrition AI-based PROTEIN mobile app for six weeks to support beneficial changes to their diet and activity levels. A food frequency questionnaire (FFQ) and an International Physical Activity Questionnaire (IPAQ) were administered. Anthropometry measurements include height, weight, body fat percentage, muscular mass, basal metabolism, and hip and waist circumferences. (B) Density plots showing inter-individual variation in host variables in response to the PROTEIN dietary intervention. The X axis represents, for each variable and each individual, the ratio of the post–pre-PROTEIN values to the pre-PROTEIN values. Different colors represent metrics belonging to various studied traits i.e., grey: alpha diversity, yellow: anthropometry, red: diet and purple: biochemistry. AI, artificial intelligence; Faith’s PD, Faith’s phylogenetic diversity; BMI, body mass index; HOMA, homeostasis model for assessment of beta cell function.

Figure 2

The impact of the PROTEIN dietary intervention on the alpha and beta diversity metrics. In the upper panel, boxplots of alpha diversity measures (A) Chao1, (B) Faith’s PD, and (C) Pielou’s evenness are shown. In the bottom panel, beta diversity is presented by PCoA, measured by non-phylogenetic (D) Bray–Curtis distance and phylogenetically aware, (E) unweighted and (F) weighted Unifrac distance metrics. Blue color corresponds to pre-PROTEIN and red color corresponds to post-PROTEIN metrics. p-Values for alpha and beta diversity metrics are from the Wilcoxon rank sum test and PERMANOVA test (999 permutations), respectively. PCoA, principal coordinate analysis; Faith’s PD, Faith’s phylogenetic diversity.

Figure 3

The associations of ASVs with the “timepoint” parameter of the MaAsLin2 multivariate regression model. The top 35 ASVs according to the significance of association are shown (p < 0.05). Associations with an FDR < 0.05 are indicated with an asterisk. Positive associations represent ASVs with increased abundance during the dietary intervention, while negative associations denote ASVs with decreased abundance. Error bars indicate standard deviation.

Figure 4

Boxplots of dietary factors significantly decreased after PROTEIN intervention. In the upper panel, daily intakes of (A) total energy, (B) carbohydrates and (C) proteins, at both timepoints, are shown. In the bottom panel, dairy intakes of food groups, (D) alcohol and beverages, (E) sweets and (F) fast food, at both timepoints, are shown. p-values were obtained from linear mixed models with participants (n = 29) as random effect, timepoint as fixed effect, and age and sex as covariates.

Figure 5

The correlation between differentially abundant genera and (A) dietary and (B) biochemistry traits at the post-PROTEIN timepoint. Significant correlations with Rho <|0.3| are not shown. Only prevalent (>50%) differentially abundant genera were tested in the correlation analysis. ⁺ p < 0.05 and * FDR < 0.05.

Figure 6

The correlation between changes (post–pre-PROTEIN) in differentially abundant genera and (A) dietary, (B) anthropometric, and (C) biochemistry traits. Only differentially abundant genera with a prevalence > 50% were tested in the correlation analysis. ⁺ p < 0.05 and * FDR < 0.05.