Research On Protein Ubiquitin Classification Algorithm Using Deep Learning

Nowadays,with the rise of artificial intelligence research,the deep learning method has become the focus of academic and industrial circles.Deep learning is the most cutting-edge and most efficient method in traditional machine learning field,and has achieved many excellent results in machine learning competitions such as ImageNet and Kaggle.This approach essentially mimics the neuronal mechanisms of the human brain in biology for feature learning.The main processes include data processing,model training,sample testing and stuff to learn the optimal parameters.Nowadays,deep learning not only is a theoretical concept,but also has penetrated into various fields and has been widely used.Such as speech recognition,face recognition,intelligent robots,intelligent diagnosis are specific applications of deep learning methods.The study of various properties of proteins is an important subject in biology,such as the prediction of classification problems of ubiquitin processes.Protein ubiquitination is a post-translational modification of proteins that are widely present in eukaryotic cells.A growing body of research has shown that ubiquitination and its reverse process deubiquitination play a key regulatory role in the process of innate and adaptive immune responses by regulating the function of different cell types in the immune system.The emergence and treatment of many major diseases affecting humans.The classification problem of protein ubiquitination is essentially a two-class problem.At present,the research on whether a protein has a ubiquitination process depends mostly on biological experiments,so it requires a lot of manpower and material resources,and takes a long time which means it is very inefficient.So far,only 1,906 proteins have been labeled as having a ubiquitination process on Uniprot,the most authoritative protein database website.Most of the existing researches based on traditional machine learning methods have made some progress,but the accuracy rate is still low.Deep learning is the most advanced and efficient machine learning method at present.Therefore,this paper deal with this protein classification problem from the perspective of deep learning methods.Moreover,in this paper,we focus on the classification of protein ubiquitination properties,and try to use protein annotation for classification prediction instead of traditional protein sequence.In addition,this paper mainly proposes two data processing methods,namely KNN-SCORE and Multi-Hot.As for model selection,this paper uses convolutional neural network,fully connected neural network and WDL model to design five models according to the training situation,and compares all models under the evaluation indexes of accuracy,true negative rate,MCC and F1 Score,the results are displayed through tables and pictures.