Machine Learning Revealed a Novel Ferroptosis-Based Classification for Diagnosis in Antiretroviral Therapy-Treated HIV Patients with Defective Immune Recovery.

Despite virological suppression, the CD4⁺ T lymphocytes are not restored in some HIV-infected patients after antiretroviral therapy. These individuals are known as immune non-responders (INRs). INRs are at high risk of developing AIDS and non-AIDS-related events and have a shorter life expectancy. Hence, it is vital to identify INRs early and prevent their complications, but there are still no specific diagnostic indicators or models. Ferroptosis has lately been reported as a type of programmed cell death, which plays an indispensable part in diverse diseases. However, its particular regulatory mechanisms remain unclear and its function in the pathogenic process of defective immunological recovery is still unknown. Blood is mainly used for rapid diagnosis because it enables quick testing. To investigate the role of ferroptosis-related genes (FRGs) in early detection of INRs, we scrutinized Gene Expression Omnibus datasets of peripheral blood samples to estimate their effectiveness. To our knowledge, for the first time, gene expression data were utilized in this study to discover six FRGs that were explicitly expressed in peripheral blood from INRs. Later on, multiple machine-supervised learning algorithms were employed, and a superlative diagnostic model for INRs was built with the random forest algorithm, which displayed satisfactory diagnostic efficiency in the training cohort (area under the curve [AUC] = 0.99) and one external validation cohort (AUC = 0.727). Our findings suggest that FRGs are implicated in the development of defective immune recovery, presenting a potential route for early detection and potential biological targets for the most effective treatment of defective immune recovery.