Integrating lipidomics and machine learning to identify plasma lipid biomarkers for MASLD-associated liver fibrosis in people with HIV

Abstract Background Liver fibrosis in metabolic dysfunction-associated steatotic liver disease (MASLD) is often unrecognized in people with HIV (PWH). Cost and operator dependency limit ultrasound and transient elastography. Altered lipid metabolism may contribute to MASLD-associated liver fibrosis in PWH. The aim was to construct and validate a plasma lipid-based diagnostic panel for MASLD-associated liver fibrosis in PWH. Methods Untargeted and targeted lipidomic profiling was performed on plasma from 380 PWH (83 with MASLD-associated liver fibrosis and 297 without) across discovery, model development, and independent validation cohorts. Results Untargeted profiling revealed 248 lipid species that differed between PWH with and without fibrosis (193 higher and 55 lower in fibrosis). Pathway analysis revealed enrichment in sphingolipid metabolism, glycerophospholipid metabolism, and insulin resistance. Targeted lipidomic profiling of the model development cohort revealed 1,673 lipid species across 33 lipid classes. Of these, 417 lipid species had correlations of |ρ| ≥ 0.3 with the controlled attenuation parameter, and 21 had correlations with liver stiffness measurement. Phosphatidylinositol and ceramide species were among those most strongly correlated with liver stiffness. Eight independent feature selection algorithms were integrated via robust rank aggregation, identifying 25 core lipid features. With the top 9 lipids, the naïve Bayes classifier gave the best discrimination among the lipid-only classifiers (area under the receiver operating characteristic curve [AUROC] = 0.83). A logistic regression model combining phosphatidylinositol PI(16:0_18:2) with alanine aminotransferase, gamma-glutamyltransferase, and the number of metabolic cardiovascular risk factors achieved an AUROC of 0.91, outperforming the lipid-only model ( P < 0.001). In an independent cohort of 99 PWH, the model’s AUROC was 0.84 versus 0.59 for fibrosis-4 and 0.63 for the aspartate aminotransferase to platelet ratio index (both P < 0.001). Conclusions Plasma lipids have diagnostic potential for MASLD-associated liver fibrosis in PWH. The combined lipid–clinical model outperformed conventional fibrosis scores and may help identify PWH who warrant transient elastography during routine HIV care.