PO15 Gene expression-based classification of atrial fibrillation and sinus rhythm across the atria

Atrial fibrillation (AF) is the most common heart arrhythmia, characterized by irregular and rapid atrial electrical activity. The literature indicates that the left and right atria differ in AF development and underlines the power of omics to uncover the complex nature of AF. This work aimed to investigate the gene expression profiles of atrial tissue from patients with AF vs sinus rhythm (SR) using data from both left and right atria. Moreover, the potential of differentially expressed genes (DEGs) to discriminate between rhythms and anatomical origin in the context of AF was explored. Data used came from the GSE79768 dataset (freely available in Gene Expression Omnibus database), which includes data from 13 patients who underwent cardiac surgery, of which 7 presented persistent AF, while the other 6 presented SR. Raw gene expression data were obtained using the GEOquery package in R. Linear model and empirical Bayes moderation were performed using the limma package in R to identify the DEGs between AF and SR groups. Genes were considered significantly differentially expressed when the adjusted p-value was lower than 0.05 and the absolute value of log2 fold change (|log2FC|) was greater than 0.5. Significant DEGs were used to create a multiclass support vector machine classifier (implemented in MATLAB) able to discriminate the rhythm (AF vs SR) and anatomical location of the sample (left atrium vs right atrium). A leave-one-out cross-validation was employed, and 10% of the DEGs were selected in each fold using the univariate ANOVA test to train the model. Before training, the data was normalized based on the training set. The performance of the model was assessed by calculating the receiver operating characteristic (ROC) curve and its area under the curve (AUC) for each class; moreover, accuracy (ACC), sensitivity (SE), and specificity (SP) were calculated considering the 0.5 operating point. The significant DEGs were 1301; 1008 upregulated genes and 293 downregulated genes. The classifier presented an overall accuracy on 81%, and good performance in all classes (AF left atrium: AUC=84%, SE=86%, SP=79%, ACC=81%; AF right atrium: AUC=77%, SE=73%, SP=95%, ACC=81%; SR left atrium: AUC=93%, SE=83%, SP= 95%, ACC=90%; AF right atrium: AUC=93%, SE=85%, SP=97%, ACC=92%). The obtained results suggest the potential use of gene expression profiles as discriminating features for AF. High classification performance for the left atrium and for SR agrees with existing research, suggesting the predominant role of the left atrium in AF maintenance and remodelling. The discrimination of AF in the right atrium, lower than in other classes, may be due to a greater heterogeneity or less distinct gene expression profiles. A key limitation of this study was the small sample size. Nevertheless, the integration of gene expression analysis with supervised classification could be a promising step toward improving AF discrimination, also integrated with other data types.