**Background:** COVID-19, caused by SARS-CoV-2, shares overlapping clinical features with other respiratory infections such as influenza, complicating accurate diagnosis. While RT-qPCR is the standard diagnostic method, it has limitations including false-negative results and inability to distinguish active infection from colonization. Host-response biomarkers in blood could serve as complementary diagnostic tools. This study aimed to identify specific blood-based gene expression biomarkers for COVID-19 diagnosis and differential diagnosis from influenza using a multi-step bioinformatics and machine learning approach.
**Methods:** Eight transcriptomic profiles of COVID-19 infected versus control samples from peripheral blood (PB), lung tissue, nasopharyngeal swab, and bronchoalveolar lavage fluid (BALF) were analyzed. For specific blood differentially expressed genes (SpeBDs), the authors identified shared pathways between PB and the three respiratory tissue sources, then extracted PB DEGs involved in those shared pathways. Nine datasets of influenza (H1N1, H3N2, B) were used to identify differential blood DEGs (DifBDs) by extracting DEGs involved in pathways enriched by SpeBDs but not by influenza DEGs. A wrapper feature selection approach supervised by four classifiers (k-NN, Random Forest, SVM, Naïve Bayes) was applied to narrow down candidate genes. Models were validated on external datasets. Performance was assessed using accuracy (ACC), area under the curve (AUC), Matthews Correlation Coefficient (MCC), sensitivity, and specificity with ten-fold cross-validation.
**Key Results:** From PB DEGs overlapping with common pathways across BALF, lung, and swab, 108 unique SpeBDs were identified. Feature selection using Random Forest selected IGKC, IGLV3-16, and SRP9 as the most predictive specific blood biomarker signatures (SpeBBSs). The Random Forest model achieved 95.92% ACC and 93.80% AUC on the feature selection dataset (GSE166190), and 93.09% ACC and 98.00% AUC on the validation dataset (Bibert et al. dataset-A). Eighty-three pathways were enriched by SpeBDs but not by any influenza strain, yielding 87 DifBDs. Feature selection using Naïve Bayes selected FMNL2, IGHV3-23, IGLV2-11, and RPL31 as differential blood biomarker signatures (DifBBSs). The Naïve Bayes model achieved >97.87% ACC and >95.00% AUC on the feature selection dataset (GSE161731-B), and 87.2% ACC on the validation dataset (Bibert et al. dataset-B).
**Clinical Implications:** This study presents a systematic strategy for identifying blood-based biomarkers that are specific to COVID-19 and can differentiate it from influenza. The identified gene panels (IGKC, IGLV3-16, SRP9 for specific diagnosis; FMNL2, IGHV3-23, IGLV2-11, RPL31 for differential diagnosis) represent minimal, clinically practical biomarker panels. These findings could facilitate development of blood-based diagnostic tests that complement RT-qPCR, potentially improving diagnostic accuracy and enabling differentiation between COVID-19 and influenza, which is critical for treatment decisions and infection control. However, further practical studies are needed to validate these combinatorial biomarkers in clinical settings.