| Abstract | We developed a novel artificial intelligence (AI) approach based on machine-learning to predict general health and food-intake parameters. This approach, named Transcriptome-driven Health-status Transversal-predictor Analysis (THTA) is relevant for markers of diabesity and is based on a non-transcriptomic, mathematics-driven approach. The prediction was based on values derived from food consumption, dietary lipids and their bioactive metabolites, peripheral blood mononuclear cell (PBMC) mRNA-based transcriptome signatures, magnetic resonance imaging (MRI), energy metabolism measurements, microbiome analyses, and baseline clinical parameters, as determined in a cohort of 72 subjects. Our novel machine learning approach incorporated transcriptome data from PBMCs as a “one-method” approach to predict 77 general health-status markers for the broad stratification of the diabesity phenotype. These markers would usually necessitate measurements using 16 different methods. The PBMC transcriptome was used to determine these 77 basic and background health markers with very high accuracy in a transversal-predictor establishment group (Pearson correlations are r = 0.98 ranging from 0.94 to 0.99). These collected variables provide valuable insides into which individual factor(s) are mainly target diabesity. Based on the “establishment group” prediction approach a further “confirmation group” prediction approach was performed, achieving a predictive potential r = 0.59 (ranging from 0.19 to 0.98) for these 77 variables. This “one-method” approach enables the simultaneous monitoring of a large number of health-status variables relevant to diabesity and may facilitate the monitoring of therapeutic and preventive strategies. In summary, this novel technique, which is based on PBMC transcriptomics from human blood, can predict a wide range of health-related markers. |
|---|