Aging-related markers in rat urine revealed by dynamic metabolic profiling using machine learning

Abstract

The process of aging and metabolism is intimately intertwined; thus, developing biomarkers related to metabolism is critical for delaying aging. However, few studies have identified reliable markers that reflect aging trajectories based on machine learning. We generated metabolomic profiles from rat urine using ultra-performance liquid chromatography/mass spectrometry. This was dynamically collected at four stages of the rat’s age (20, 50, 75, and 100 weeks) for both the training and test groups. Partial least squares-discriminant analysis score plots revealed a perfect separation trajectory in one direction with increasing age in the training and test groups. We further screened 25 aging-related biomarkers through the combination of four algorithms (VIP, time-series, LASSO, and SVM-RFE) in the training group. They were validated in the test group with an area under the curve of 1. Finally, six metabolites, known or novel aging-related markers, were identified, including epinephrine, glutarylcarnitine, L-kynurenine, taurine, 3-hydroxydodecanedioic acid, and N-acetylcitrulline. We also found that, except for N-acetylcitrulline ( p < 0.05), the identified aging-related metabolites did not differ between tumor-free and tumor-bearing rats at 100 weeks ( p > 0.05). Our findings reveal the metabolic trajectories of aging and provide novel biomarkers as potential therapeutic antiaging targets.

Causal associations of genetically predicted gut microbiota and blood metabolites with inflammatory states and risk of infections: a Mendelian randomization analysis

“…In line with us, C5DC levels were positively correlated with the inflammatory marker IL-1β (Guerreiro et al, 2021). Increased C5DC levels are linked to cardiovascular risks (Zhao et al, 2020) and aging (Carlsson et al, 2021), both of which are commonly accompanied by prolonged chronic inflammation (Shi et al, 2021). Notably, C5DC elevations caused by glutaryl-CoA dehydrogenase deficiency are generally seen in the inborn glutaric aciduria type I disorder, leading to neurological dysfunction and high inflammatory states (Zhao et al, 2014).…”

Serum Metabolomic Profiling in Aging Mice Using Liquid Chromatography—Mass Spectrometry

“…Machine learning can improve diagnosis and treatment, transforming biomedicine and the medical practice [ 15 ]. In recent studies, machine learning has been used to discover genetic and metabolite markers linked to aging [ 30 , 31 , 32 , 33 ]. We selected 20 candidate metabolites related to aging using four algorithms, including LR, GNB, SVM, and RF.…”

Recent advances in LC‐MS‐based metabolomics for clinical biomarker discovery

“…Machine learning has been applied to analyze and interpret massive and complex data, and is able to discover markers with better predictive power compared to biostatistical models. Supervised learning algorithms, including support vector machines, random forests, and neural networks, can learn and map LC‐MS data to their corresponding subgroups, such as a patient group and a healthy control group (Leavell et al, 2020; Liebal et al, 2020) and have positive impacts on the identification of biomarkers for nonalcoholic steatohepatitis and fibrosis (Perakakis et al, 2019), chronic kidney (Y. Guo et al, 2019), and aging (Shi et al, 2021). By combining a panel of marker peaks, ensemble methods (e.g., random forest) can improve the predictive power of the model and have been used to identify predictive features for type 2 diabetes (Grissa et al, 2020).…”