OBPred: feature-fusion-based deep neural network classifier for odorant-binding protein prediction

Odorant-binding proteins (OBPs) are small, water-soluble, globular, and ligand-specific proteins found in the sensillar lymph in insects and mucus layer in vertebrates. They function as a carrier for transferring odorants to their respective olfactory receptors. However, the functionality and expression of OBPs are not limited to olfaction mechanism and olfactory organs. OBPs are involved in various other functions like measuring odorant concentration and deactivating odorants. They are found in insect taste organs, venom glands, and sex hormone glands. To gain more insight into the functioning and expression pattern of OBPs, it has become essential to identify OBP sequences. The available OBP prediction methods are primarily based on sequence similarity, which seems inefficient since OBPs share low interspecies and intraspecies sequence similarities. Effective representation of protein sequence-based features is essential for the development of predictive models. Single feature representation like physicochemical properties and secondary structure content (PCSS), amino acid composition (AAC), and sequence graph transform may be insufficient. Therefore, we propose two feature combination schemes: feature fusion and feature concatenation, for OBP prediction. The role of conventional machine learning algorithms and deep learning algorithms have not been explored much for identifying OBP sequences. The performance metrics suggest the feature fusion scheme enhances the overall performance of the OBP prediction classifiers. Deep networks yield better accuracy as compared to traditional Machine learning algorithms. The performance of the deep neural network and one-dimensional convolution neural network was found to be comparative. Among all feature representation schemes, the feature fusion scheme with AAC + PCSS gave the best performance. The accuracy of deep neural network and convolutional neural network-1D for AAC + PCSS representation was found to be 98.6% and 96.5%, respectively. We propose OBPred, a deep neural network model with the fused ACC + PCSS feature set, for OBP identification. The current study is the first attempt, to our knowledge, to predict OBPs using deep neural networks.