Abstract

Long non-coding ribonucleic acids (lncRNAs) are sequences that do not encode proteins and have been identified as essential regulators of ischemic stroke. Stroke is a leading cause of serious long-term disability, and several lncRNAs have recently been discovered to influence ischemic stroke processes, particularly the immune response following a stroke. This study focuses on the expression profiles of long non-coding RNAs and microRNAs in ischemic stroke samples. The phases of differential expression analysis include quality checks, generating transcript-level counts, computing gene-level counts, and performing differential expression analysis using a statistical Wald test implemented in DESeq2. The gene expression counts generated are then used to classify samples into their respective groups with the help of machine learning algorithms. After comparing acute ischemic stroke to control samples, the study findings identified 2,341 genes that were up-regulated and 2,087 genes that were down-regulated. In the sub-acute sample compared to the control sample, 3,639 genes were up-regulated and 3,479 genes were down-regulated. Furthermore, comparisons among the three groups—acute, sub-acute, and control—revealed that 53 genes were up-regulated and 10 genes were down-regulated. The accuracy of the Random Forest, Decision Tree, and MLPClassifier models was around 89%, surpassing the logistic regression and SVM models, which achieved accuracies of approximately 80% and 87%, respectively. The Random Forest model completed in 0.015 seconds, while the Decision Tree model took 0.018 seconds. Therefore, it can be concluded that Random Forest provides an improved classification model for classifying samples into their respective groups.

Key words: Downregulated genes, machine learning, principal component analysis, upregulated genes, support vector machine.