Nonlinear dimensionality reduction of mass spectrometry data for odor sensing

In this paper, we propose a neural network based dimensionality reduction approach for mass spectrometry data. Since a mass spectrum of chemical molecule is considered to be highly related to its characteristics of smell, its low-dimensional representation can be used as a proper feature for odor sensing application such as e-nose. We designed a nonlinear autoencoder with three hidden layers and applied it to a data set of mass spectra of odorant molecule. It was found that our method is able to compress the data with less reconstruction error than that of linear transformation such as PCA. Moreover, compressed information might be more suitable for odor sensing application.