RT-PCR accuracy improvement for SARS-CoV-2 detection using deep neural networks

Abstract

Analysis of fluorescence-based Real-Time Polymerize Chain Reaction (RT-PCR) amplification data is increasingly used to detect multiple pathogens and variants of those rapidly and simultaneously through gene expression. If the gene of interest for the pathogen exists in the sample, then the PCR amplification data forms a type of logistic curve (sigmoid) with an exponential phase. If the pathogen does not exist in the sample, then the amplification signal produces just a noise. The traditional approach for RT-PCR data analysis focuses on the value of the point where the cutoff-threshold (CT) line crosses the exponential phase of the curve if it exists. Focusing on the determination of the CT value too often causes mislabeling of pathogens as either false positives or false negatives. Therefore, this research demonstrates the possibility of improving the accuracy of RT-PCR pathogen identification performance.