Computational Prediction of Aptamer-Target Interaction Based on Deep Learning Approaches

Abstract

Abstract

Background: Aptamers are short oligonucleotides (DNA/RNA) or peptide molecules that can selectively bind to their specific targets with high specificity and affinity. As a powerful new class of amino acid ligands, aptamers have high potential in bio sensing, therapeutic and diagnostic fields. Aim: In this study, we present AptaNet - a new deep neural network- for prediction of the aptamer-protein interaction pairs by integrating features derived from both aptamers and the target proteins. Methods: Aptamers were encoded by using two different strategies including k-mer and revck-mer frequency. Pseudo amino acid composition (PseAAC) was applied to represent target information by using of 24 physicochemical and conformational properties of proteins. To handle the imbalance problem in the data, we applied neighborhood cleaning algorithm. Our predictor was constructed based on a deep neural network and an optimal features were selected using random forest algorithm. Results: As a result, 99.79% accuracy were achieved for the training dataset, and 91.38% accuracy were obtained for the testing dataset. AptaNet achieved high performance on our constructed aptamer-protein benchmark dataset. Also, a set of optimal features were selected, which significantly contributed to the predictions of aptamer-protein pair interactions. Conclusion: The results indicate that AptaNet can be a helpful tool for identifying novel aptamer-protein interacting pairs and building more-efficient insights into the relationship between aptamers and proteins. Moreover, the optimal features analyzed in this study can provide useful insights into the mechanisms of aptamer-target interactions, and aptamer scientists might take advantage of these computational techniques to develop more accurate and more sensitive aptamers.