Blanca Lacruz-Pleguezuelos

This study is part of the AI4Food project, which aims to identify the physiological and behavioral factors influencing weight variation in overweight and obese individuals through integrative analysis of multi-source data. A 30-day crossover intervention was conducted on 93 participants at IMDEA Nutrition Institute, combining a traditional phase (questionnaires, food diaries) and a digital phase (Fitbit Sense Smartwatch, FreeStyle Libre 2 glucose sensor). The collected data encompassed sleep, glucose, skin temperature, nutrition, and gut microbiome information. After rigorous preprocessing and alignment by intervention days, unsupervised clustering, association analysis, and machine learning models were applied to uncover physiological patterns linked to weight dynamics. Results showed greater weight reduction during the digital phase in Group 2, while between-group differences were not significant. Significant associations were observed between nocturnal respiratory rate, skin temperature, and weight variation. Under participant-level five-fold cross-validation, Random Forest models achieved mean accuracies of 64.3% for thermoregulation and 61.7% for glucose phenotype prediction, with RMSSD, restlessness, and nocturnal respiratory rate among the most influential predictors. In contrast, microbiome composition remained almost stable over the short intervention period. These findings support the potential of AI-driven integration of wearable and nutritional data to stratify metabolic phenotypes and advance personalized nutrition strategies in obesity management.

Obesity is a heterogeneous condition comprising a continuum of phenotypes with various metabolic and inflammatory profiles. Metabolically healthy obesity (MHO) identifies individuals with obesity but a relatively preserved metabolic state, although little is known about the gut microbiome features underlying this phenotype. Here, we analyzed gut microbial network structures of 931 individuals living with metabolically healthy non-obesity (MHNO), MHO, metabolically unhealthy non-obesity (MUNO), and metabolically unhealthy obesity (MUO), performing cross-sectional analyses on feces shotgun metagenomics data. Individuals with MHNO and MHO harbor more robust and functionally cohesive microbial networks, while communities from MUO and MUNO phenotypes exhibit a potentially dysbiotic state with reduced connectivity. A nutritional intervention cohort showed an improvement in network connectivity in parallel with metabolic improvements. Our findings show differences in microbial connectivity and association patterns across metabolic and obesity phenotypes, shedding light on how distinct microbial network structures may associate with host metabolic health and disease.

This study aimed to develop an artificial intelligence (AI) algorithm capable of distinguishing Alzheimer’s disease (AD) from healthy patients using gut microbiome metagenomics data. To do so, 16S rRNA gene and Whole Genome Shotgun (WGS) datasets available in the literature were utilised. Data was pre-processed and filtered. Then, an initial analysis was done to study classical parameters of microbiome data, such as alpha and beta diversity, as well as to unveil taxonomical differences between the groups that might be influencing disease development, using a unified workflow to enable integration of 16S rRNA gene and WGS data. Neural Network algorithms were subsequently applied to develop the classifier. First a classical multi-layer perceptron (MLPNN) architecture was tested, which showed limited classification performance, particularly in detecting AD cases. These results were improved using a convolutional neural network (CNN) architecture, due to its better comprehension of hierarchic data like taxa relative abundances sorted following phylogenetic tree structure. In addition, these AI methods were compared with a random forest (RF) classifier, a traditional machine learning algorithm. All models struggled to accurately identify AD cases due to the low number of samples used for algorithm training. Although the RF showed a better performance under such circumstances, observing the evaluation metrics the application of AI to this task reveals promising upon higher amounts of training data. The use of SMOTE, a data augmentation approach, confirmed this assumption improving performance across all models but specifically in the CNN. These results support the utility of microbiome-based diagnostics in AD, but highlight the need for larger, diverse datasets and multi-omics approaches to improve model reliability and uncover disease mechanisms.

Fungi represent less than 1% of the gut microbiota; However, their importance in host homeostasis and disease is increasingly recognized. Accurate characterization of the gut mycobiome from metagenomic data remains a significant challenge due to the low abundance of fungal DNA, the performance of bacteria-oriented classifiers, and the limited availability of curated fungal reference databases. To overcome these issues, we developed FungiGutDB v1.0, a curated database containing 304 taxa previously identified in culture-dependent human studies, and we integrated the database in a reproducible workflow to ease its application (FungiGut). Benchmarking analyses demonstrated that FungiGut achieved a substantially lower false positive rate in mock communities compared to standard non-gut-specific fungi databases. When applied to real metagenomic datasets, FungiGut successfully characterized the gut mycobiome, identifying Saccharomyces cerevisiae as the predominant species in healthy individuals, along with common dietary fungi found in fermented dairy products (Penicillium camemberti, Debaryomyces hansenii, Kluyveromyces lactis, Pichia kudriavzevii). In contrast, samples from patients with non-responsive celiac disease showed a higher relative abundance of opportunistic pathogens and a lower number of diet-associated taxa, suggesting a trend toward a dysbiotic mycobiome profile. By limiting classification to fungal species previously isolated from the human gut, FungiGut minimizes misclassifications derived from environmental or plant-associated taxa, which often lead to mistaken interpretation of the results. Overall, FungiGut offers a biologically consistent and reproducible approach to gut mycobiome profiling, improving taxonomic accuracy and strengthening confidence in the interpretation of fungal metagenomic data in human microbiome research.

Obesity is a heterogeneous condition comprising a continuum of phenotypes with various metabolic and inflammatory profiles. Metabolically healthy obesity (MHO) identifies individuals with obesity but a relatively preserved metabolic state. However, the criteria defining MHO remain inconsistent, and little is known about the gut microbiome (GM) features underlying this intermediate phenotype. Here, we aim to describe microbial structures contributing to metabolic health and disease. To do so, we analyzed the GM of 959 individuals classified as metabolically healthy non-obese (MHNO), MHO, metabolically unhealthy non-obese (MUNO), and metabolically unhealthy obese (MUO), using stool shotgun metagenomics. MHO subjects display intermediate anthropometric and biochemical profiles, with a GM composition and diversity in an in-between state among MHNO and MUO individuals. Network science analyses reveal that metabolic health, rather than obesity, drives microbial connectivity: MHNO and MHO individuals harbor more robust and functionally cohesive microbial networks, whose most influential nodes are focused toward SCFA production. In contrast, MUO and MUNO communities exhibit a dysbiotic state with reduced connectivity and increased influence of low-abundance, ectopic and potentially pro-inflammatory species resulting in a damaged, unstable microbial community network. These findings suggest that metabolic disorders disrupt microbial ecology beyond compositional shifts, emphasizing the need for systems-level approaches. Our findings show differences in microbial connectivity and association patterns across metabolic and obesity phenotypes, shedding light on how distinct microbial structures may contribute to metabolic health and disease.

Non-responsive celiac disease (NRCD) poses a challenge for clinicians due to the persistence of symptoms despite maintaining a gluten-free diet (GFD). This study investigated the gut microbiome, mucosal integrity, and metabolomic profiles of 39 NRCD patients to gain insights into the underlying mechanisms contributing to symptom persistence. Two distinct clusters of patients were identified based on clinical and demographic variables not influenced by gluten consumption. Cluster 1, labelled “Low-grade symptoms,” displayed milder symptoms and lower inflammatory markers. In contrast, Cluster 2, named “High-grade symptoms,” exhibited more severe gastrointestinal and extraintestinal symptoms, along with elevated inflammatory markers and increased intestinal permeability. Despite similar mucosal damage in both clusters, network analysis of the gut microbiome revealed specific microbial taxa with potential functional implications. Cluster 1 displayed a microbiome associated with immune homeostasis and gut barrier integrity, potentially lowering inflammation and symptom severity. In contrast, Cluster 2 had a distinct microbiome linked to lactate production, Th17 activation, possibly contributing to heightened inflammation and gastrointestinal symptoms. Metabolomic analysis revealed differential metabolites between clusters, particularly in amino acid metabolism pathways. Metabolites associated with specific symptoms were identified, implicating their potential role in symptom manifestation. Notably, vitamin D deficiency was observed in both clusters, suggesting its relevance in the context of NRCD. The study highlights the importance of the gut microbiome, mucosal integrity, and metabolic pathways in symptom persistence among NRCD patients. The associations between microbial-derived metabolites and symptom severity provide valuable insights into potential therapeutic targets. Further research is needed to validate these findings and develop targeted interventions for improving clinical outcomes in NRCD patients.

The consumption of highly processed foods, along with other dietary and lifestyle poor habits, has an impact on health by increasing the risk of several non-communicable pathologies, such as diabetes. Gut microbiome composition, in specific, can be modulated by nutrients, deriving in different metabolic outcomes that have an influence on this high disease susceptibility and making it a possible therapeutic target for these comorbidities. In this work, gut microbiome of 60 and 46 individuals from 2 different studies focused, among other aspects, on diet-microbiome interactions, was characterised. By means of differential abundance analyses and supervised machine learning techniques based on random forest, gradient boosting and support vector machines, a set of microbial genera that could be potential biomarkers for the differentiation of individuals with poorer dietary patterns was discovered, after comparing coincidences in these taxa among classifiers and testing them for significant differences. Among these, Dialister , Phocea and Pseudoflavonifractor were suggested to have a role in the way highly processed foods affect health negatively, along with Prevotellaceae NK3B31 group and an undetermined genus from Muribaculaceae in the opposite sense. Furthermore, all the identified genera in this study had already been linked to type 2 diabetes, among which Bacteroides and Pseudoflavonifractor proved to be differentially abundant in groups of individuals with different levels of biomarkers for this disease. Nevertheless, further research via longitudinal studies and experimental validation of these genera should be carried out to confirm the association of these taxa with diet and diabetes.

Food–drug interactions (FDIs) occur when a food item alters the pharmacokinetics or pharmacodynamics of a drug. FDIs can be clinically relevant, as they can hamper or enhance the therapeutic effects of a drug and impact both their efficacy and their safety. However, knowledge of FDIs in clinical practice is limited. This is partially due to the lack of resources focused on FDIs. Here, we describe FooDrugs, a database that centralizes FDI knowledge retrieved from two different approaches: a natural processing language pipeline that extracts potential FDIs from scientific documents and clinical trials and a molecular similarity approach based on the comparison of gene expression alterations caused by foods and drugs. FooDrugs database stores a total of 3 430 062 potential FDIs, with 1 108 429 retrieved from scientific documents and 2 321 633 inferred from molecular data. This resource aims to provide researchers and clinicians with a centralized repository for potential FDI information that is free and easy to use. Database URL: https://zenodo.org/records/8192515 Database DOI: https://doi.org/10.5281/zenodo.6638469

The increasing prevalence of diet-related diseases calls for an improvement in nutritional advice. Personalized nutrition aims to solve this problem by adapting dietary and lifestyle guidelines to the unique circumstances of each individual. With the latest advances in technology and data science, researchers can now automatically collect and analyze large amounts of data from a variety of sources, including wearable and smart devices. By combining these diverse data, more comprehensive insights of the human body and its diseases can be achieved. However, there are still major challenges to overcome, including the need for more robust data and standardization of methodologies for better subject monitoring and assessment. Here, we present the AI4Food database (AI4FoodDB), which gathers data from a nutritional weight loss intervention monitoring 100 overweight and obese participants during 1 month. Data acquisition involved manual traditional approaches, novel digital methods and the collection of biological samples, obtaining: (i) biological samples at the beginning and the end of the intervention, (ii) anthropometric measurements every 2 weeks, (iii) lifestyle and nutritional questionnaires at two different time points and (iv) continuous digital measurements for 2 weeks. To the best of our knowledge, AI4FoodDB is the first public database that centralizes food images, wearable sensors, validated questionnaires and biological samples from the same intervention. AI4FoodDB thus has immense potential for fostering the advancement of automatic and novel artificial intelligence techniques in the field of personalized care. Moreover, the collected information will yield valuable insights into the relationships between different variables and health outcomes, allowing researchers to generate and test new hypotheses, identify novel biomarkers and digital endpoints, and explore how different lifestyle, biological and digital factors impact health. The aim of this article is to describe the datasets included in AI4FoodDB and to outline the potential that they hold for precision health research. Database URL https://github.com/AI4Food/AI4FoodDB