Specificity of SARS-CoV-2 Real-Time PCR Improved by Deep Learning Analysis

Area of Science:

• Molecular Biology
• Bioinformatics
• Machine Learning

Background:

• Real-time PCR (RT-PCR) is crucial for pathogen diagnosis, relying on threshold cycle (C) value estimation.
• Current C value models use approximations that neglect stochastic fluorescence variations, impacting accuracy with large sample sizes (e.g., SARS-CoV-2 pandemic).

Purpose of the Study:

• To develop and validate a novel, C value-independent method for RT-PCR data interpretation.
• To introduce qPCRdeepNet, a deep learning model for analyzing RT-PCR fluorescence signals.
• To assess qPCRdeepNet's utility as a real-time quality assurance tool for RT-PCR results.

Main Methods:

• Developed and trained a deep learning model, qPCRdeepNet, to analyze raw fluorescence data from RT-PCR.
• Evaluated the model's performance using the TaqPath COVID19 Combo Kit assay for SARS-CoV-2 detection.
• Compared the deep learning approach against traditional C value-based interpretation.

Main Results:

• qPCRdeepNet successfully interprets RT-PCR data independently of the C value.
• The model demonstrates robust performance, particularly for SARS-CoV-2 detection assays.
• The deep learning approach provides a reliable alternative for real-time monitoring and interpretation of RT-PCR results.

Conclusions:

• The qPCRdeepNet model offers a novel and potentially superior method for RT-PCR data analysis.
• This C value-independent approach can enhance diagnostic accuracy and serve as a quality assurance tool.
• The model has broad applicability for various RT-PCR assays, potentially replacing the conventional C value paradigm.