Classification of single cells by Raman spectroscopy and machine learning: comparison of common algorithms

Surface-enhanced Raman spectroscopy (SERS) with in situ or colloidal produced metallic nanoparticles is a powerful tool for fast and reliable measurement of pro- and eukaryotic cells. Recently, much research has been focused on the use of a wide range of machine learning methods to detect minute differences in the spectra of various groups of cells. One aspect which is often neglected in current literature is the No Free Lunch Theorem, which states that one cannot assume a priori that a specific method is appropriate for a given task. Accordingly, we compare a wide range of commonly used linear and nonlinear classification methods applied to three biological datasets (prokaryotic cells, cyanobacteria, and Hodgkin lymphoma cells) and assess their performance and interpretability. Results demonstrate that while a pre-selection based on the size of the dataset and the complexity of the classification task at hand, comparing several algorithms is still crucial to achieving optimal performance. For this purpose, the Python tool developed herein was made to be easily adaptable for different data and classification methods.