Unveiling lipidomic signatures: Leveraging machine learning to predict obese and non-obese status

Background: Obesity is a multifactorial disease associated with dysfunctional adipose tissue. The prompt identification of metabolic alterations and biomarkers associated with increased risk of obesity is crucial for preventing obesity-related complications. Aim: This study aims to develop a machine learning model that establishes lipidomic profiles as biomarkers to classify individuals as either obese or non-obese accurately. Hypothesis: We hypothesize that lipidomic profiles can serve as a reliable biomarker for classifying the obese and non-obese status of individuals. METHOD: We performed a cross-sectional predictive study on 122 individuals: obese (n=75) and non-obese controls (n=47) categorized based on body mass index (BMI). We used mass spectrometry for untargeted lipid profiling of adipose tissue samples from both groups. Multivariate analysis was performed, followed by building predictive models using the Random Forest and K-Nearest Neighbor (KNN) algorithms, with lipid abundance values as features for classification. Result: A total of 595 lipid species were identified through untargeted lipidomic analysis. Decision-based random forest model (Out of bag, OBB error: 3.28%) revealed Ceramide(Cer), Diacylglycerol(DG), Phosphatidylethanolamine(PE), Sphingomyelin(SM), and Phosphatidylglycerol(PG) as important lipid biomarkers to distinguish non-obese from obese participants. Among features identified by the random forest model, PE 15:0_18:1 and PG 15:0_18:1 were identified as differential biomarkers for obesity due to their elevated levels in obese participants (Wilcoxon, p < 0.05). The KNN model (neighbors = 13) achieved a classification accuracy of 87%, with an AUC value of 0.86, indicating an ideal model for differentiating non-obese and obese individuals based on lipid profiles. Conclusion: Collectively, these findings demonstrate that lipid profiling, integrated with machine learning, provides an effective approach for classifying obese and non-obese individuals, highlighting specific lipids as potential indicators of obesity. This project was funded by the National Institute of Health grant number R01HL161386 (PI AMM) and by UIC's Strategic Initiative, 2023 (PI AMM) This abstract was presented at the American Physiology Summit 2025 and is only available in HTML format. There is no downloadable file or PDF version. The Physiology editorial board was not involved in the peer review process.