Study On Protein Interactions Coding And Protein Interactions Prediction Models

A series of important life activities,such as apoptosis,immune response and metabolic pathway,are achieved through the interactions between proteins.Protein-proteins interactions help to elucidate the molecular mechanisms of life activities and have a certain role in promoting disease treatment and new drug development.With the advent of the proteome era,enormous amounts of proteomic data have been accumulated.However,how to deal with these massive data,how to extract the interactions between proteins from a large number of protein sequence data,and how to construct a network of interactions between proteins have become an urgent problem to be solved in current proteomics research.Therefore,this dissertation takes protein-proteins interactions as the research goal,and studies the coding methods of proteins sequences and protein-proteins interactions prediction based on amino acid sequences.The main research contents and results of the thesis are as follows:(1)To improve the predictive performances of protein-proteins interactions,this dissertation is based on Deep Neural Network(DNN)combined with Conjoint Triads(CT),Auto-Covariance(AC),Local Descriptor(Local Descriptor,LD)three protein coding methods,constructing DNN-CT,DNN-AC and DNN-LD three kinds of protein-proteins interactions prediction models.The dropout was used to optimize the prediction performances of three protein-proteins interactions.The results showed that the accuracies of the DNN-CT,DNN-AC and DNN-LD three methods increased from 97.11%to 98.12%,96.84%to 98.17%,and 95.30%to 95.60%,respectively.The loss values of DNN-CT,DNN-AC and DNN-LD decreased from 27.47%to 14.96%,65.91%to 17.82%,and 36.23%to 15.34%,respectively.These results show that dropout can improve the accuracy of the prediction model and reduce the loss rate of the prediction model,which provides a feasible solution for the optimization of protein interaction prediction model.(2)Based on CT and AC5 a new feature coding method is proposed:Conjoint Triad Auto Covariance(CTAC).This method combines the CTAC coding method with machine learning algorithms such as deep neural network,support vector machine,adaptive lifting algorithm and random forest to construct different protein-proteins interaction prediction models.The experimental results show that the protein interaction prediction model based on CTAC coding method can obtain better prediction performance in Benchmark dataset,meanwhile,they also can obtain better prediction results on four external datasets respectively.Compared with the existing methods,the experimental results are superior.Existing protein sequence coding methods.(3)Aiming at the shortcomings of existing protein sequence coding methods CT,AC and LD without considering the sequence relationships of the whole amino acid sequence,a new protein sequence coding method based on Matrix of Sequence(MOS)was proposed.The coding method considers the sequence relationship of the entire amino acid sequence,on the other hand,it reduces the dimension of the vector space,reduces the calculation amount,and improves the training speed.Based on DNN and MOS,the protein-proteins interactions prediction model DNN-MOS was constructed.The experimental results showed that the prediction performances of DNN-MOS were better than most existing prediction methods,and it can be used as a useful supplement for protein-proteins interactions prediction,which provided a new solution for protein-proteins interactions prediction.In view of the shortcomings of existing coding methods,this dissertation proposes two new methods of protein coding.Based on deep learning and traditional machine learning combined with multiple coding methods,a number of protein interaction prediction models were constructed,and the model was optimized by dropout,which provided technical support for protein interaction prediction.