Fluorescence Spectral Shape Analysis For Covid-19 Virus Identification: Machine Learning Approach

Fast and reliable tests for the new coronavirus are urgently needed. Current Polymerase Chain Reaction based virus detection approaches are typically time-consuming and expensive. Technologies capable of providing a fast, real-time and non-contact detection of virus contamination and real-time virus classification are not yet available. Here, we demonstrate the potential of a fluorescence detection technique along with machine-learning based classification for virus detection. The ultraviolet (UV) light irradiated virus emits a fluorescent signal with a characteristic spectrum, which is regarded as a fingerprint for the virus. We analyzed eight virus samples including a heat-inactivated SARS-CoV-2 (virus causing COVID-19) and collected a number of emission spectra. Machine learning techniques are applied to discriminate among the candidate viruses via classifying a number of spectra data collected. First, Principle Component Analysis (PCA) was applied to reduce spectra data dimensionality. Then support vector machine (SVM) with various kernel functions (kernel-SVM), k-nearest-neighbor (k-NN) and Artificial Neural Networks (ANN) methods were used to classify these viruses with dimension-reduced data from PCA. We found that dimension-reduced data in 3 principal components (PCs) space performs better than that in 2 PCs space in the machine learning algorithms mentioned above. Variance ratio analysis is able to explain nearly 95% of variance which allows nearly 100% accuracy of predictions for 25% data test set randomly chosen from the whole dataset. Finally, cross validation (CV) analysis is applied to kernel-SVM and k-NN methods.