Optimizing genomic signal extraction of COVID-19 variants using the multi linear predictive coding (M-LPC) method

Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) is a novel coronavirus transmitted to humans and causes COVID-19. In some cases, COVID-19 shows symptoms similar to other illnesses, such as influenza, making it difficult to diagnose. We propose a machine learning-based feature extraction method called Multi Linear Predictive Coding (M-LPC) to classify COVID-19 genome signals. M-LPC employs the LPC principle method along with a sliding window technique. This approach utilizes mathematical calculations to generate basic statistical values for identifying features within DNA sequences, including nucleotide frequency, GC distribution, as well as maximum, minimum, mean, and standard deviation. These features can accurately distinguish COVID-19 variants from viruses that have similar symptoms, such as influenza. The advantage of the M-LPC method lies in its ability to be applied to DNA sequences of any length that have been converted into genomic signals, and generate simple features for subsequent machine learning-based classification. Our research shows that M-LPC successfully extracts essential features of the COVID-19 genome signal with an accuracy of 99.86% for three different disease classes and 92.45% for 12 different classes. Therefore, our proposed method with high accuracy can serve as a decision support tool for more accurate diagnosis of COVID-19."