Prediction of molecular substructure using mass spectral data based on deep learning

Abstract

The success of machine learning algorithms generally depends on effective features. Prediction of molecular substructure based on mass spectral data is try to extract more useful information or feature expression. In this paper, deep learning (DBN) was used to extract mass spectral features automatically. A large dataset consisting 11 molecular substructure is extracted from NIST mass spectral library. The experimental results show that deep learning (DBN) achieve best classification performance in 11 molecular substructure contrasting traditional classification methods.