Research On Drug Response Prediction Algorithm Based On Matrix Factorization

Cancer has always threatened human life and health,and providing personalized medical services for cancer patients has become the main trend in cancer treatment in the future.Using biometric data to accurately predict the response level of cancer cell lines and drugs,and to formulate precise treatment plans for patients,is the key to personalized medicine.After in-depth investigation and research,this paper uses the two databases of Cell Line Encyclopedia(CCLE)and Cancer Drug Sensitivity Genomics(GDSC)to integrate the characteristic information provided by drug response observation data,gene expression data and medicinal chemical structure data to design Two drug response prediction models-based on the K-nearest neighbor constraint matrix factorization model and the Bayesian matrix factorization model based on the social trust relationship,experimental results show that the algorithm can accurately predict the unknown drug response score.Based on the K-nearest-neighbor constraint matrix factorization model,the Knearest neighbor constraint is introduced on the basis of the similarity-constrained matrix factorization model.The model uses K-nearest neighbor selection to filter out the noise information in the similarity network,and sets different features according to the level of similarity the weight makes the model more reasonable.The results show that,compared with the matrix factorization model with similarity constraints,the Knearest neighbor constraint matrix factorization model has a higher average Pearson correlation coefficient of 0.02 and a lower average root-mean-square error on the 23 drugs of CCLE;The average Pearson correlation coefficient between the predicted value and the observed value on the 135 drugs in GDSC is higher than 0.02,and the average root mean square error is lower than 0.36.It shows that K-nearest neighbors help improve the accuracy of drug response prediction.The algorithm also proved that both the cell line similarity network and the drug similarity network can help improve the accuracy of drug response prediction,but the contribution of the cell line similarity network in the K-nearest neighbor constraint matrix factorization model is greater than that of the drug similarity network.The Bayesian matrix factorization model based on social trust relationship is to make up for the shortcomings of ordinary low-rank matrix factorization.Based on the Bayesian matrix factorization model of social trust relationship,the similarity network is integrated into the Bayesian matrix factorization framework in a way different from regularization and factorization.The similarity network is used to construct social relationships and the trust relationship judgment mechanism is introduced.The hyperparameters of the model are sampled according to the social trust relationship.Compared with the BPMF model,the sampling of parameters and hyperparameters is more reasonable,which improves the prediction accuracy of drug response without increasing the complexity of the model.Compared with the K-nearest neighbor constraint matrix factorization model,the principle design of the Bayesian matrix factorization model of social trust relations closely relies on the Bayesian probability theory,which has good interpretability and also obtains the same good prediction accuracy.In addition,the results also prove that the Bayesian matrix decomposition model of social trust relationship has better anti-noise ability,lower risk of overfitting,and less impact of data sparsity.In this paper,two drug response prediction models are designed in response to different problems in drug response research.The research results prove the effectiveness of the model designed in this paper,the prediction accuracy has been effectively improved,and new research conclusions have been obtained,which have certain reference value and research significance for cancer treatment and drug response research.This article has 23 figures,9 tables,and 101 references.