Prediction of growth/no growth status of previously unseen bacterial strain using Raman spectroscopy and machine learning
Takashi Yamamoto, J. Nicholas Taylor, Shige Koseki, Kento Koyama
Abstract
We develop a method to predict the growth/no growth response of previously unseen bacteria, using Raman spectral features and a machine-learning model. Twenty-one strains of bacteria were isolated from seven commercially available fresh-cut vegetables. Twenty Raman spectra of single cells were acquired for each isolated strain. The growth/no growth responses of each strain in a liquid medium were evaluated with two levels of sodium acetate concentrations, two incubation temperatures, and eight sampling times using optical density to confirm growth limit conditions. The Raman spectra of 20 strains and their forty-eight growth/no growth responses were used to train an artificial neural network model to predict the growth/no growth of unknown bacteria. Our model predicted the growth/no growth of 21 unknown bacteria with an overall accuracy of 90%. Such rapid characterization of unknown bacterial growth using Raman spectroscopy will be valuable for the food manufacturing industry, where unknown bacteria are often encountered.