Integrating Bioinformatics and Machine Learning Algorithms to Screen Inflammatory Biomarkers for Atrial Fibrillation and Experimental Validation

Objective To identify potential biomarkers of atrial fibrillation (AF) using bioinformatics, machine learning (ML) and experimental methods.

Methods AF transcriptomic data were obtained from the GEO and MSigDB databases to screen for differentially expressed genes (DEGs) and inflammation-related gene sets (IRGs). The overlap between DEGs and IRGs was used to identify the DE-IRGs set. Two machine learning algorithms were used to filter AF-related DE-IRGs. Twelve male SD rats were randomly divided into a control group and an AF group. Rats received acetylcholine (66 μg/mL) and calcium chloride (10 mg/kg) via tail vein injection for five consecutive weeks to establish the AF model. Morphological characteristics of atrial myocytes were assessed with HE staining, while RT-qPCR and immunohistochemistry (IHC) were used to measure changes in mRNA and protein levels.

Results In the training set, 119 DEGs were identified, with IRGs showing the highest correlation with two co-expression modules (r = 0.60 and 0.56, P < 0.0001). The intersection of DEGs and IRGs yielded nine DE-IRGs. The SVM-RFE and RF algorithms identified 5-hydroxytryptamine receptor 2B gene (HTR2B), latent-transforming growth factor beta-binding protein 2 gene (LTBP2), matrix remodeling associated protein 5 gene ( MXRA5), and transforming growth factor β-induced protein gene (TGFBI) as highly expressed in both the training and test sets in AF groups with high diagnostic efficacy (AUC > 0.77). The electrocardiogram limb leads of AF rats showed numerous f-waves. HE staining revealed disorganized atrial myocyte arrangement and inflammatory cell infiltration in the AF group. The TUNEL fluorescence assay showed an apoptosis rate of (55.34 ± 4.29)% in the AF group, compared to (8.69 ± 3.12)% in the control group (P = 0.0001). There were statistically significant differences in the relative expression levels of LTBP2 and TGFBI between the AF group (4.97 ± 4.20, 2.62 ± 1.85) and the control group (1.12 ± 0.21, 1.18 ± 0.77) (P = 0.0137, P = 0.0444). IHC revealed positive rates of LTBP2 and TGFBI proteins in the AF group of (36.50 ± 1.31)% and (27.39 ± 4.57)%, respectively, compared to (22.95 ± 2.62)% and (18.26 ± 3.70)% in the control group (P = 0.0008, P = 0.0485).

Conclusion The characteristic genes LTBP2 and TGFBI are highly expressed in AF and serve as valuable diagnostic biomarkers.