Immune and inflammatory features associated with pyroptosis- and myocardial fibrosis-related genes in HFpEF: a machine learning study

Abstract

Heart failure with preserved ejection fraction (HFpEF) is a heterogeneous syndrome marked by inflammation, immune dysregulation, and myocardial fibrosis. Whether pyroptosis intersects with fibrotic remodeling in HFpEF remains uncertain. Two murine cardiac transcriptomic datasets were integrated after batch correction and intersected with curated pyroptosis- and myocardial fibrosis-related gene sets. Functional enrichment, univariate logistic regression, support vector machine–recursive feature elimination (SVM-RFE), and least absolute shrinkage and selection operator (LASSO) regression were used for feature selection. Complete separation in the ordinary three-gene logistic fit prompted refitting with Firth bias-reduced logistic regression. Performance was assessed by apparent area under the curve (AUC), stratified bootstrap optimism correction, and repeated nested cross-validation; the panel was re-estimated in a human left-ventricular dataset for exploratory replication. Cell-type Identification by Estimating Relative Subsets of RNA Transcripts (CIBERSORT) estimated immune-cell fractions, and expression was examined in a high-fat diet (HFD)/N ω -nitro-L-arginine methyl ester hydrochloride (L-NAME) murine model and whole blood from patients with HFpEF. Six candidates— Pparg , Ace2 , Smad2 , Lcn2 , Ptpn11 , and Sqstm1 —were identified. Gene Ontology and Kyoto Encyclopedia of Genes and Genomes analyses linked them to SMAD binding, autolysosome/amphisome activity, cellular senescence, efferocytosis, and renin-angiotensin signaling. In the 45-sample discovery cohort, the fixed Pparg - Ptpn11 - Ace2 Firth model had an apparent AUC of 1.0000; bootstrap correction yielded an AUC of 0.9971 (95% percentile interval, 0.9780–1.0000). Repeated nested cross-validation produced a mean out-of-fold AUC of 0.9286 (2.5%–97.5% percentile interval, 0.8726–0.9700), revealing additional feature-selection uncertainty. Re-estimation in the human dataset represented panel-level replication rather than validation of a locked cross-species model. No inferred immune-cell fraction differed between HFpEF and control after false-discovery-rate correction, although exploratory correlations were observed among deconvolved fractions and with Ptpn11 expression. HFD/L-NAME-treated mice developed diastolic dysfunction, reduced exercise capacity, pulmonary congestion, cardiomyocyte hypertrophy, and myocardial fibrosis despite preserved systolic function; myocardial Ptpn11 , Pparg , and Ace2 transcripts and their proteins were reduced. In patients with HFpEF, whole-blood PTPN11 mRNA was lower than in controls and inversely associated with N-terminal pro-B-type natriuretic peptide and C-reactive protein; these unadjusted findings remain exploratory. The three-gene panel is hypothesis-generating rather than a stable diagnostic model; its transportability, selection stability, and clinical relevance require prospective validation.