Multi-omics integration and machine learning for the identification of molecular markers of insulin resistance in prepubertal and pubertal children with obesity

Abstract

Childhood obesity can develop early in life leading to the appearance of metabolic alterations such as insulin resistance (IR). If maintained during adulthood, obesity and IR usually derive into the development of more serious conditions like Type II Diabetes and cardiovascular disease, which considerably increase morbidity and mortality in affected populations. As many other complex disorders, obesity and its associated cardiometabolic comorbidities constitute a complex phenotype arising from the interaction between an at-risk molecular profile (involving genomics, transcriptomics, epigenomics and proteomics disturbances) and environmental exposures. On this sense, one of the most promising fields of research in obesity has involved the identification of early-life predictive molecular biomarkers able to stratify patients according to their risk for developing cardiometabolic complications later in life. Interestingly, the ideal and most robust biomarker discovery approach would involve the simultaneous analysis of multiple omics data layers at a once, allowing tracking a molecular disturbance from all its possible dimensions. Due to the complexity of omics data, nevertheless, new and innovative analytics approaches have been demanded. In the middle of this need, bioinformatics and artificial intelligence (AI) have experienced a remarkable boost due to their ability to automatically obtain descriptive or predictive models from massive amounts of data (Big Data). The present Doctoral Thesis gathers a series of research works in which bioinformatics and AI are conveniently applied to several obesity observational omics research projects for identifying new molecular biomarkers of IR and metabolic alterations in children and adolescents with obesity. Study populations are composed of more than 2000 Spanish children with ages ranging from 2-18 years. In summary, the results presented in the present Doctoral Thesis indicate that; 1) obesity is a complex disorder resulting from the interaction between genetic and environmental factors, 2) the creation of predictive tools based on the combination of small-risk effects genetic variants is an interesting but simple strategy for predicting future obesity, 3) multi-omics research approaches in obesity are necessary to understand the complex molecular mechanisms underlying disease, and 4) the application of eXplainable Artificial Intelligence (XAI) machine learning (ML) models can help us to unravel the complex relationships between omics molecular elements. The application of multi-omics research approaches and the use of complex analytical tools (such as bioinformatics and AI) are the correct way for approaching a true implementation of a personalized care in obesity. Further studies like those presented in the present Doctoral Thesis and as well as larger cohorts projects should be encouraged in order to validate presented findings. This will require a close collaboration between clinicians and basic researchers, and the creation of multidisciplinary teams, in which the presence of mixed bioinformatics profiles will be of great importance.