The Study On Age Estimation Of Deep Vein Thrombosis Based On Plasma Protein Expression Profiles

Objective:Agilent protein chip technology combined with four kinds of machine learning algorithms was used to detect and analyze the changes of plasma protein profile in the process of deep vein thrombosis(DVT)in rats,to explore the regularity of temporal changes of plasma protein expression profile in thrombosis,and to infer the age of thrombosis.Method:The rat model of deep vein thrombosis was established and the rats were divided into control group and operation group.The control group was not treated(n = 10).The operation group included 1 h、6 h、12 h、1 d、2 d、3 d、4 d、5 d、6 d、7 d、10 d、14 d and 21 d after thrombosis(n = 10).The serum samples of 140 rats were collected,and the protein of plasma was detected by Agilent protein chip technology.Support vector machine model,logistic regression model,random forest model and Fisher linear discriminant model are combined to analyze the sample data.The importance of sample variables was ranked by the optimal mathematical model.Result:(1)The control group showed the change of blood stasis and blood stasis,and no thrombosis.A small amount of platelets were observed after 12 hours.A large amount of platelets were observed after 1 day.A small amount of fibroblast was observed after 3 days,the density of thrombosis was increased.The number of fibroblasts was increased after 5days,the exudation of fibrin-like substance was also increased.After 14 days,the cellulose like substance accumulated;(2)The protein expression profiles of plasma samples were different at different time,and the peak position and content were different;(3)Principal component analysis and orthogonal partial least squares discriminant analysis showed that there were significant differences among the sample groups;(4)The classification prediction model of multiple thrombosis time was established.The accuracy of support vector machine model was 86.8% and ROC value was 0.97,logistic regression model was 63.1% and ROC value was 0.95,random forest model was 86.8% and ROC value was 0.98,Fisher linear discriminant model was 81.5% and ROC value was 0.94.Comprehensive analysis showed that random forest model was more suitable for the experimental data,the external validation accuracy was 85.7%;(5)The importance of sample variables was sorted by using the random forest model.The results showed that protein peak P11 and protein peak P10 were the most important.The protein peak P11 with molecular weight of 56.9 KDa and P10 with molecular weight of 45.8KDa may be closely related to thrombosis by literature review and molecular prediction.Conclusion:The protein chip technology can quickly detect and accurately obtain the protein expression profiles of plasma samples at different thrombosis time points in rats.Combined with the machine learning algorithms,we can analyze the temporal changes of plasma protein expression profiles,and establish four mathematical algorithm models.The random forest model has the highest accuracy,the accuracy was 85.7%,which provides a new method and idea for the inference of thrombosis time.