Proposal of Data Augmentation Method for Metabolome Estimation from Raman Spectral

Abstract

Generally, in deep learning of image recognition, there is an established data augmentation method to increase the training data. However, there is no established data augmentation method that can handle numerical data analyzed in experiments, etc. Therefore, we propose a method for data augmentation of such numerical data. We are developing a method to analyze serum components using a Raman spectrometer and to estimate the metabolome and other components in serum from Raman spectral data. Serum samples are samples obtained through animal experiments, and the number of samples is limited. Although the learning accuracy of deep learning decreased when the number of samples was small, the proposed data augmentation method improved the learning accuracy.