SARS-CoV-2 Screening on the Multiplex Real-Time RT-PCR Gene Cycle Threshold - A Machine Learning Approach

Abstract

Early diagnosis of the lethal SARS-CoV-2 virus determines a patient’s survival rate. Highly transmissible novel coronavirus prevention is possible with effective, rapid diagnostic strategies. The reverse transcription polymerase chain reaction (RT-PCR), a globally adopted SARS-CoV-2 detection technique, provides better diagnosis results. The output of the RT-PCR test produces the amplified gene scores of ORF1a/b, S, N, E and RdRp. This study intends to evaluate the performance of the RT-PCR-based COVID-19 diagnosis using machine learning models. The confirmatory genes ORF1b, E and RdRp and their cycle threshold (Ct) values are the main parameters used to build the machine learning model for SARS-CoV-2 screening. The real-time dataset collected from the Indian Council of Medical Research (ICMR) database containing missing, redundant information is processed and eliminated. Statistical interpretations are performed with demographic information to understand the dynamics of the disease prevalence in India. Binary classification models delivered promising results in discriminating the samples of two classes. The models were examined further to scrutinize their performance via evaluation metrics such as balanced accuracy, f1-score, ROC curve, precision and recall. This algorithmic assessment exhibits a better outcome on the RT-PCR-based SARS-CoV-2 disease diagnosis.