Recently, researchers have shown an increased interest in studying volatile metabolite as disease biomarker which can provide non-invasive, safer and cheaper diagnosis tools. Volatile metabolites are also being produced during secondary metabolite and they are not directly involved in normal growth or reproduction of cell but crucial for the survival in the environment. These metabolites play vital role in defensing plants from herbivores, attracting pollinators, seed dispersers and also plant-plant interaction. It is important to investigate the role of volatile metabolites in human healthcare and chemical ecology, however a limited amount of metabolite data that contain in the existed database become a big challenge for researchers to determine biological activities of these volatile metabolites. In previous study, we have accumulated 341 volatiles emitted by biological species associated with 11 types of biological activities and deposited the data into our database, which is called KNApSAcK Metabolite Ecology Database. In this study, we have extended the database by accumulating more data with total of 619 volatiles with 22 different biological activities. 12 models of Deep Neural Network (DNN) were developed to predict the biological activities of volatile metabolites using H2O package in R software. As a comparison, the performance of deep neural network was also being compared with Random Forest (RF) and Gradient Boosting Machine (GBM). Based on the experimental results, DNN6 show the best performance with lower MSE (0.2746) and highest accuracy (76.71%) compared to other models in performance of cross validation. Thus, it indicates that DNN has best accuracy than others. Therefore, DNN method is recommended to be used in the prediction of biological activities of the volatile metabolites. The prediction outcome may be useful for the discovery of novel agricultural tools and biomarkers in medical diagnostic field.
