Machine learning approach for label-free rapid detection and identification of virus using Raman spectra

Objective

The objective of this study was to develop a robust method for rapid detection and identification of the virus based on Raman spectroscopy combined with machine learning approach.

Methods

We have used saliva spiked with different bacterial viruses such as P1 Phage, M13 Phage, and Lambda Phage, for demonstrating the utility of this method for virus detection. The Raman spectra collected from a large number of independent samples, each of different phages with and without saliva were used to train a supervised convolutional neural network (CNN) with its hyperparameters optimized by Bayesian optimization. The CNN method was not only able to detect the presence of a phage but was also able to identify the phage type using unprocessed Raman spectra having high noise. In addition, a semi-supervised auto-encoder was utilized for differentiating healthy saliva from saliva spiked with phages thereby making it possible to detect the presence of phages in saliva samples.

Results

The CNN could identify the virus with an accuracy of 98.86% based on ten-fold cross-validation, precision of 98.8%, recall of 98.7%, and F1 score of 98.7%. The area under the curve of receiver operating characteristic curve was 0.99. Autoencoder was capable of differentiating healthy saliva from the virus spiked saliva with an accuracy of 99.7% in a semi-supervised manner. Thus, Raman spectroscopy coupled with machine learning approach was able to directly detect and identify the virus without consuming time for lengthy sample processing.

Conclusion

A robust method based on Raman spectroscopy coupled with machine learning may be capable of detection and identification of the virus even from the signal with low intensity and high noise. This label-free method is fast, sensitive, specific, and cost effective.