Robustness of FTIR-Based Ultrarapid COVID-19 Diagnosis Using PLS-DA

Area of Science:

• Virology
• Medical Diagnostics
• Data Science

Background:

• The Omicron variant of SARS-CoV-2, responsible for COVID-19, presents challenges due to high transmissibility despite vaccination efforts.
• Current diagnostic methods for COVID-19 face limitations during widespread outbreaks.
• Chemometric approaches offer potential for rapid, cost-efficient, and non-destructive disease diagnosis.

Purpose of the Study:

• To systematically evaluate the robustness of chemometric diagnosis models for COVID-19.
• To compare the performance of models trained on smaller versus larger datasets, including augmented data.
• To establish reliable diagnostic tools for emerging SARS-CoV-2 variants.

Main Methods:

• Utilized chemometrics for developing diagnostic models based on spectral data.
• Employed Monte Carlo cross-validation and permutation tests to assess model robustness.
• Trained and validated models using both smaller (real and augmented) and larger (augmented) datasets.

Main Results:

• Diagnosis models trained on larger datasets showed superior accuracy and statistical significance.
• Achieved excellent performance metrics: Q² > 99% and AUROC = 100% for models using larger datasets.
• Demonstrated the robustness and reliability of chemometric diagnosis for COVID-19.

Conclusions:

• Larger datasets significantly enhance the accuracy and robustness of chemometric COVID-19 diagnosis models.
• Chemometric methods provide a promising avenue for rapid and reliable detection of SARS-CoV-2 infections.
• The validated models are crucial for effective patient isolation and management during the evolving pandemic.