Periodic Reporting for period 1 - NUTRIclock (NUTRIclock: a tool based on microbiome and artificial intelligence for implementing precision nutrition.)
Période du rapport: 2023-06-01 au 2025-09-30
Aging is the decline of physiological properties that affect all organisms. It is an unavoidable process and constitutes the greatest risk factor for the development of chronic diseases like neurodegenerative disorders or cancer. According to the WHO, in 2020 the number of people over 60 years old overcame those under 5 years, and the number of people over 80 years old will triple between 2020-2050. Also, this demographic shift is not a high-income country problem exclusively as low- and middle-income countries are also experiencing these changes. Thus, this increase in life expectancy makes it extremely urgent to understand and develop mechanisms to promote healthy aging. Nutrition has traditionally been a non-pharmacological mean to influence human physiology and has demonstrated effectiveness at slowing aging progression via different processes like modulating metabolism, lowering inflammation and/or influencing on the gut microbiome. The gut microbiome that comprises all gut microorganisms, is strongly influenced by nutrition and has shown capable to interact and modulate these aging-related mechanisms as well. The NUTRIclock project proposes the generation of a “nutritional clock”, an aging clock influenced by microbiome and nutrition profiles to evaluate how their changes may modulate aging process. This clock will calculate the biological age of a subject based on this data in contrast to the chronological age. To achieve that, we proposed combining advanced Artificial Intelligence (AI) algorithms known as Neural Networks (NNs) with metagenomic and nutritional data. The NNs provide great efficiency handling large and complex datasets as well as extracting patterns from non-linear data correlations. Thus, NUTRIclock project will provide a tool to evaluate the effect of nutritional interventions on the biological age of a patient and therefore their effectiveness, helping to promote healthy aging. Also, due to NNs capacity to manage large datasets, this will allow to process higher amounts of information per patient. Facilitating the further development of tailored nutritional recommendations for a specific individual profile and advancing in the personalized nutrition field.
During the 2 years of the MSC action, I worked on developing NUTRIclock. First, the available metagenomics and metabolomics datasets useful for its development were identified. Since it’s rare to find both types of datasets for the same volunteer group in human studies, the nutrients’ relative abundances were inferred from metagenomic data. The metagenomic data source used was the R package CuratedMetagenomicDataset, which contains metagenomic datasets processed using the Metaphlan3 pipeline to ensure consistency. Data was filtered, preprocessed and studied preliminary using different R packages (e.g. MMUPHin, MaAsLin2, MelonnPan). The final samples obtained belong to 3,664 healthy patients, aged 18 and over, assessing the batch effect among different studies and retaining bacterial data present in at least 10% of the samples and with a minimum relative abundance of 1e-5. This data revealed an increase in alpha diversity with age but no changes in beta diversity. Metabolites information was inferred using microbiome data and aging-related microbiome species along with metabolites were identified.
Next, different architectures and combinations of data were tested to obtain the best balance between age prediction and performance of the algorithm. The following architectures were used: I) Multilayer perceptron NN (MLPNN), a classical NN approach with hidden layers extracting information from input data. II) Convolutional NN (CNN), specific for images treatment. III) Variational autoencoders (VAE), architectures that extract information of the input data generating a latent representation that might be used to reconstruct the original data. These architectures were used on microbiome relative abundances, inferred metabolites relative abundances and/or metabolites-related pathway abundances. The architecture with the better prediction/performance balance was a CNN + VAE on microbiome data combined with MLPNN + VAE on inferred metabolites data.
Finally, the NUTRIclock algorithm was tested on samples from a diseased population and from a nutritional intervention study. I) The algorithm showed increased prediction of biological age in patients with age-related diseases like diabetes and cancer. This was the expected result, whereas they are considered “older”, and confirmed that the algorithm works. The effect was particularly increasing with the severity of the conditions. II) When applied to a nutritional intervention study in overweight/obese patients, it showed a discreet rise in biological age for non-responders and no changes for responders, aligning with previous results given the short duration of the intervention (3 weeks) and mild responder criteria (only 2kg weight loss).
Next, different architectures and combinations of data were tested to obtain the best balance between age prediction and performance of the algorithm. The following architectures were used: I) Multilayer perceptron NN (MLPNN), a classical NN approach with hidden layers extracting information from input data. II) Convolutional NN (CNN), specific for images treatment. III) Variational autoencoders (VAE), architectures that extract information of the input data generating a latent representation that might be used to reconstruct the original data. These architectures were used on microbiome relative abundances, inferred metabolites relative abundances and/or metabolites-related pathway abundances. The architecture with the better prediction/performance balance was a CNN + VAE on microbiome data combined with MLPNN + VAE on inferred metabolites data.
Finally, the NUTRIclock algorithm was tested on samples from a diseased population and from a nutritional intervention study. I) The algorithm showed increased prediction of biological age in patients with age-related diseases like diabetes and cancer. This was the expected result, whereas they are considered “older”, and confirmed that the algorithm works. The effect was particularly increasing with the severity of the conditions. II) When applied to a nutritional intervention study in overweight/obese patients, it showed a discreet rise in biological age for non-responders and no changes for responders, aligning with previous results given the short duration of the intervention (3 weeks) and mild responder criteria (only 2kg weight loss).
NUTRIclock represents a new approach to the use of AI algorithms in the study of microbiome, nutrition and aging. Some previous studies explored the use of simpler architectures of Neural Networks for this purpose. But this is the first time that more complex and diverse NN architectures were used and that metabolites information were also added for the prediction. I am now working on testing a promising architecture for this purpose, the so known as Graphical NNs. This architecture may fit better gut microbiome communities as they interpret information from community network interactions. An improvement in NUTRIclock’s performance is anticipated. Also, a scientific publication is also under preparation to cover all the results and development of NUTRIclock. In addition, I obtained a software product in TLR3 (proof of concept stage) that could be of potential interest to different stakeholders as it will facilitate the evaluation of health status based on “easy” to obtain data (in comparison to other methods) as microbiome composition. Thus, both researchers and nutritionists involved in the precision and personalized nutrition field as well as food industry agents with interest in determining the effect of their products on human health, may be potential users of the final algorithm developed. Regarding this, IMDEA Nutrition is already negotiating with a spin-off of the center the licensing for the future use and commercialization of the algorithm generated under NUTRIclock project.
Nevertheless, the MSC action allowed me to grow as an independent researcher and had great impact on my scientific career at different levels: I) Receiving training in NNs use and development together with leadership training, in prestigious courses from the MIT (Massachusetts), EMBO and ESADE business school. II) Collaborating and building an international network of partners, producing a review on the use of NNs on metagenomic data (corresponding authorship). III) Participating in international congresses with oral (ISMB Montreal 2024, Beneficial Microbes Conference 2025) and poster presentations (EFFOST Brugge 2024). IV) Mentoring my own master students, one of them currently my first PhD student. V) Being awarded with a renowned contract in Spain to become a fully independent PI (Ramón y Cajal contract).
Nevertheless, the MSC action allowed me to grow as an independent researcher and had great impact on my scientific career at different levels: I) Receiving training in NNs use and development together with leadership training, in prestigious courses from the MIT (Massachusetts), EMBO and ESADE business school. II) Collaborating and building an international network of partners, producing a review on the use of NNs on metagenomic data (corresponding authorship). III) Participating in international congresses with oral (ISMB Montreal 2024, Beneficial Microbes Conference 2025) and poster presentations (EFFOST Brugge 2024). IV) Mentoring my own master students, one of them currently my first PhD student. V) Being awarded with a renowned contract in Spain to become a fully independent PI (Ramón y Cajal contract).