Conditional Generative Adversarial Network for Spectral Recovery to Accelerate Single-Cell Raman Spectroscopic Analysis

Abstract

Raman spectroscopy is a powerful tool to investigate cellular heterogeneity. However, Raman spectra for single-cell analysis are hindered by a low signal-to-noise ratio (SNR). Here, we demonstrate a simple and reliable spectral recovery conditional generative adversarial network (SRGAN). SRGAN reduced the data acquisition time by 1 order of magnitude (i.e., 30 vs 3 s) by improving the SNR by a factor of ∼6. We classified five major foodborne bacteria based on single-cell Raman spectra to further evaluate the performance of SRGAN. Spectra processed using SRGAN achieved an identification accuracy of 94.9%, compared to 60.5% using unprocessed Raman spectra. SRGAN can accelerate spectral collection to improve the throughput of Raman spectroscopy and enable real-time monitoring of single living cells.