Classification of COVID19 Patients Using Robust Logistic Regression.

Area of Science:

• Genomics
• Infectious Diseases
• Biostatistics

Background:

• The Coronavirus disease 2019 (COVID-19) pandemic, caused by the Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), impacts global health.
• SARS-CoV-2 is thought to enter the human body through the airway epithelium, triggering a host response.
• Gene expression levels in the upper airway may indicate SARS-CoV-2 infection, but data can be affected by contamination and labeling errors.

Purpose of the Study:

• To develop a robust computational model for COVID-19 diagnosis using gene expression data.
• To identify genes significantly associated with COVID-19 and predict disease status simultaneously.
• To address data quality issues like contamination and labeling errors in gene expression datasets.

Main Methods:

• Regularized logistic regression model was employed as a classifier for COVID-19 diagnosis.
• Density power divergence robust estimating methods were utilized to handle data contamination and labeling errors.
• The performance of the robust method was compared against classical maximum likelihood estimators with LASSO and adaptive LASSO penalties.

Main Results:

• The proposed robust logistic regression model demonstrated stability and reliability in the presence of data errors.
• The model successfully identified genes related to COVID-19 and predicted disease cases based on their expression levels.
• Comparison indicated the robustness of the density power divergence method over traditional estimators in this context.

Conclusions:

• Robust statistical methods are crucial for accurate COVID-19 diagnosis from potentially erroneous gene expression data.
• The developed regularized logistic regression model offers a reliable approach for identifying COVID-19 biomarkers and predicting infection.
• This approach can improve the accuracy of diagnostic tools in real-world clinical settings where data quality may vary.