Matrix Completion For Prediction Of The Antigenicity Of Influenza Viruses

Influenza is an acute respiratory infection with strong infectivity and fast propagation speed,which is caused by influenza virus.And influenza prevention has been highly valued by all countries of the world.At present,hemagglutination inhibition test(HI)is one of the conventional means to obtain influenza virus antigenic characteristics.However,the interference and restriction of the experimental conditions will cause inaccuracy or missing data from the HI datasets.In the work,a novel low-rank matrix completion algorithm was proposed to obtain reliable and complete HI datasets.In the method,the similarities of proteins were integrated into matrix completion algorithm,which is referred to as biological matrix completion with side information(BMCSI).The BMCSI model was applied to HI datasets of H3N2 influenza virus from 1968 to 2003 to predict the antigenicity of influenza viruses.Firstly,low react values in the data were re-evaluated and missing values were recovered.The root-mean-square error(RMSE)is adapted to evaluate the effectiveness of the prediction method.Compared to previous methods,the RMSE in the method is reduced by 37% in a 10-fold cross validation analysis.Based on the cartographies constructed from imputed data,we showed that the antigenic evolution of H3N2 seasonal influenza is S-shaped while the genetic evolution is half-circle shaped.The spearman correlation between genetic and antigenic distances(among antigenic clusters)is 0.83,demonstrating a globally consistency and some local discrepancies between influenza genetic and antigenic evolution.Finally,a simple method was proposed to optimize vaccine strains from antigenic cartography and evaluate previous vaccine strain selections.In addition,the model was also applied to the datasets of 491 cancer cell lines' responses to 24 chemical compounds from CCLE.This framework uses both the correlations among rows and columns of the matrix and the power of gene expression features in predicting cancerous drug sensitivity.And then,the genomic correlated of drug sensitivity would be found using the datasets.