Machine learning to improve the interpretation of intercalating dye-based quantitative PCR results

Area of Science:

• Medical Diagnostics
• Computational Biology
• Infectious Diseases

Background:

• Mucormycosis diagnosis relies on complex interpretation of intercalating dye-based quantitative PCR (IDqPCR) signals.
• Visual assessment of IDqPCR curves can be subjective and requires specialized expertise.
• Developing objective and accessible diagnostic tools for mucormycosis is crucial.

Purpose of the Study:

• To evaluate the efficacy of a Machine Learning (ML) approach for interpreting IDqPCR signals in mucormycosis diagnosis.
• To enhance the accuracy and accessibility of mucormycosis diagnosis using computational methods.
• To develop a user-friendly tool for non-specialists to interpret IDqPCR results.

Main Methods:

• Applied ML classification to 734 IDqPCR results (74 positive, 660 negative for mucormycosis).
• Extracted 14 features from amplification and denaturation curves, combined with clinical data.
• Trained and evaluated 48 meta-classifiers using majority voting on an external dataset.

Main Results:

• ML meta-classifiers achieved Kappa coefficients > 0.83 on the external dataset.
• Six meta-classifiers demonstrated perfect classification accuracy (Kappa = 1).
• The ML approach significantly improved upon traditional visual reading of IDqPCR curves.

Conclusions:

• The proposed ML-based approach provides rigorous interpretation of IDqPCR curves for mucormycosis diagnosis.
• This method makes accurate mucormycosis diagnosis accessible to non-specialists in molecular diagnostics.
• A free online application is available for classifying IDqPCR data.