SARS-CoV-2 Diagnosis Using Transcriptome Data: A Machine Learning Approach.

Area of Science:

• Bioinformatics
• Genomics
• Machine Learning

Background:

• The global SARS-CoV-2 pandemic necessitates accurate and timely identification of infected individuals.
• Current diagnostic methods like PCR and antigen tests have limitations, including potential false negatives.
• Effective control of COVID-19 spread relies on precise detection to isolate cases and prevent transmission.

Purpose of the Study:

• To develop a highly accurate machine learning classification model for distinguishing between COVID-19 positive and negative individuals.
• To leverage transcriptome data and gene expression profiles for improved diagnostic accuracy.
• To explore various feature selection algorithms and classification models for optimal performance.

Main Methods:

• Utilized transcriptome data from SARS-CoV-2 patients and healthy controls.
• Applied three distinct feature selection algorithms to identify relevant genes.
• Trained and evaluated seven different machine learning classification models.
• Incorporated analysis of differentially expressed genes (DEGs) between the groups.

Main Results:

• The study identified mutual information and differentially expressed genes (DEGs) as effective features for classification.
• The combination of mutual information (or DEGs) with Naïve Bayes (or Support Vector Machine) classifiers yielded the highest accuracy.
• Achieved a classification accuracy of 0.98 ± 0.04, significantly outperforming existing methods.

Conclusions:

• Machine learning models, particularly those utilizing gene expression data, offer a promising avenue for accurate COVID-19 diagnosis.
• The developed model demonstrates potential for reliable identification of SARS-CoV-2 infection, aiding in pandemic control efforts.
• Further validation and clinical implementation of this approach could enhance diagnostic capabilities.