| The problem of classification and prediction of proteins is a relatively important issue in bioinformatics.Taking ubiquitination of proteins as an example,this is a process of specific modification that is importantly associated with various types of life activities.Ubiquitination plays an important role in the localization,metabolism,regulation and degradation of proteins.Traditional biological experiments that in order to determine ubiquitinated proteins and detect ubiquitination sites require a lot of manpower and material resources.If other methods can be used to screen out proteins that may have ubiquitination property then biological experiments followed,then it is easy to save a lot of cost and generate great value.The current related research is mainly based on the traditional machine learning algorithms.Most of research focus on the prediction of ubiquitination sites rather than judge whether the unknown protein can be ubiquitinated.Besides,there are still some low-accuracy,unreasonable problems.At the same time,such algorithms require a large amount of additional attribute information that is artificially labeled.These algorithms cannot be applied to those proteins that are completely new or whose properties are incomplete.Deep learning is an important branch of machine learning,and neural network model is one of the main models in deep learning.This type of method is currently the most significant way to do research on medical and bioinformatics,and has already made great breakthroughs in many related fields.In this paper,positive and negative samples are selected from a large number of protein sequence samples,then the positive samples are expanded.The sequence of the protein has been segmented into equal length parts according to the nature of the protein by bioinformatics related tools.Without affecting the prediction of unknown proteins,ubiquitination site information is added to the label.The sequence data is encoded according to one hot encoding as well as AAindex protein information library.Based on the particularity of such problems and the unprecedented achievements of deep learning,this paper designs a classification prediction model using only protein sequence information based on convolutional neural networks.In addition,a prediction algorithm based on recurrent neural network is implemented and compared with the CNN model proposed in this paper.After implementing various data processing methods and deep learning models,this paper proposes a general procedure for protein sequence classification.When deal with other protein properties,just need to make a few simple changes to the input data,then the model can be used to solve new problems.Finally,a protein ubiquitination online classification prediction web platform has been designed to take full advantage of CNN model prediction results.The platform was optimized based on possible usage scenarios and related stress tests were conducted.Compared with the traditional machine learning methods,the complexity of the data has been significantly reduced in the specific field of protein ubiquitination prediction,so it is simpler to process data.When each model uses its own full amount of data,the accuracy results of the proposed model and the best SVM algorithm are really close,which also means the proposed mode is superior to any other models.When the amount of data is close,the proposed method is better than traditional machine learning methods in terms of accuracy and other evaluation index. |
