Document Type : Research Paper

Authors

• Nikta Gohari Sadr ¹
• Armin Behjati ²
• Fatemeh Zare-Mirakabad ¹

¹ Department of mathematics and computer science, Amirkabir University of technology

² Department of mathematics and computer science, Amirkabir university of technology

Abstract

RNA-protein interactions play essential roles in many biological processes, such as gene regulation and fundamental cellular processes related to human, animal, and plant diseases. However, the patterns of these interactions are not fully understood. The experimental methods to solve this problem are expensive and time-consuming. Therefore, there is a compelling need for developing reliable computational methods. Predicting these interactions requires structural information about RNA and protein, which is not always available. On the other hand, results of the research on transformers show that they can efficiently extract biochemical, biophysical, and structural features from molecule sequences. In this experiment, we use ProtAlbert and DNABERT transformers to provide a good representation for RNA and protein sequences. Then we feed the feature vectors to an asymmetric Siamese network to predict whether they interact with each other or not. The experimental results indicate that our method achieves superior performance with an average accuracy of 92.3% and an average area under the curve of 96.6%.

Keywords

• DNABERT
• Deep Learning
• ProtAlbert

Main Subjects

• Bioinformatics