Prediction Of Protein-ATP Binding Sites Based On Deep Learning

ATP is an indispensable substance that provides energy to cells to ensure that various biological life activities can be carried out in an orderly manner.The interaction between ATP and protein is achieved by the combination of ATP and ATP binding sites on proteins,Protein is the power provided by ATP to perform various biological functions.Therefore,it is possible to accurately predict the protein-ATP binding site,which is of great significance to our study of protein function and other biological fields.With the continuous development of computer science,deep learning has gradually entered people's attention and become a hot field.There are many links in the original data that we can't detect but are very important,and there are also some useless information.Therefore,it is found that the internal correlation in the data establishes a form that can compactly express the original data,which will also improve the prediction result.We use the convolutional auto-encoder to extract important associations in the data and make a new compact representation of the data.Due to its good performance,many prediction problems with loci are predicted by convolutional neural networks,in this paper we use a convolutional neural network that combines multi-dimensional convolution kernels as our classifier.We establish a deep learning prediction model consisting of an auto-encoder and a convolutional neural network.This prediction model adopts a two-dimensional data format on the input form of the data,and the final category label is obtained through the prediction model of the deep learning.In this paper,some work has been done one feature selection.Some biochemical attributes have been selected to improve the prediction accuracy.It is also proved that the selected features are closely related to the research questions in this paper.Through several sets of experiments and comparison with other existing methods,we can see that the prediction model of this paper has good prediction performance.The accuracy of the method on the independent verification set reached 96.5%,and the MCC reached 0.516,which was excellent in performance in all aspects.From the experimental results,it can be proved that the prediction model of this paper has better prediction accuracy and good performance in solving the problem of protein-ATP binding site prediction.