Exploring potential gene signatures in dengue through machine learning and deep learning approaches

Dengue is a major public health problem that affects millions of people globally. The present study used microarray data to identify differentially expressed genes (DEGs) during dengue clinical conditions. The microarray datasets GSE84331, GSE18090, GSE43777, and E-MTAB-3162 were downloaded and analyzed using statistical analysis (Unpaired t-test). This was followed by Machine Learning (ML) and Deep Learning (DL) techniques with recursive feature elimination and genetic algorithms implemented to identify the potential biomarkers. Further, functional enrichment, platelet signaling, and protein-protein interaction (PPI) network analysis were performed to explore the potential diagnostic markers associated with dengue. Among all ML/DL models, the Random Forest algorithm outperformed on baseline data and identified 27 DEGs in the dengue fever (DF) vs. control (C) group and 13 DEGs in filtered data of the severe dengue (SD) vs. DF group. Likewise, the Support Vector Machine with Genetic Algorithm (SVM-GA) hybrid model outperformed the SD vs. C group and identified 79 DEGs. Based on the analysis, the study identified seven hub genes such as PIK3R1, GATA3, ZFPM, SKAP1 (involved in hemostasis, platelet activation, aggregation, and production), TP63, ZBTB20, and ZEB2 (abnormal hard palate morphology) for dengue diagnosis. Further, the hub genes may facilitate the development of reliable diagnostic potential; their prognostic utility requires further validation in larger, more diverse cohorts.