Integrating Machine Learning and Single-Cell Sequencing: Decoding Histone Deacetylase Genes-Based Diagnostic Features and Immune Cell Dynamics in Ischemic Stroke

Abstract Ischemic stroke (IS) involves dysregulated histone deacetylase (HDAC)-mediated epigenetic pathways, a mechanism often implicated in microglial activity during stroke. However, the clinical relevance and transcriptional mechanisms of HDACs in human IS remain unclear. Using GEO datasets (GSE22255 and GSE58294), we analyzed HDAC expression and identified significant downregulation of HDAC2 and HDAC4 and upregulation of HDAC1, HDAC6, and HDAC8 in IS patients. We developed machine learning models for stroke prediction, among which the random forest (RF) model demonstrated superior accuracy and lower residuals over support vector machine (SVM). Five hub genes (HDAC8, HDAC6, HDAC2, HDAC1, HDAC4) were selected to construct a nomogram for individualized risk assessment, which exhibited robust predictive performance and clinical utility. Consensus clustering revealed two distinct IS subtypes: cluster A with elevated HDAC1/2/8 and high immune infiltration, and cluster B with increased HDAC4. Functional enrichment highlighted oxidative stress and inflammatory pathways. Immune analyses revealed subtype-specific HDAC-immune correlations, and single-cell RNA sequencing showed prominent HDAC1/2 expression in glial cells and macrophages, with microglia playing a key role. The upregulation of the five HDAC genes was validated by qRT-PCR in a MCAO model. Our study delineates the expression profile, prognostic value, and immune-modulatory role of HDACs in IS, providing a predictive model and molecular subtyping framework to guide future therapeutics.