The Construction and Evaluation of an Age Prediction Model Using Deep Neural Network Technique
Objective To screen age-related differentially expressed genes (DEG)of healthy cases in GEO database,and to construct and evaluate an age prediction model based on the deep neural networks (DNN). Methods We collected the transcriptome data of 583 healthy cases from GEO database.We screened DEG by using the"limma"package in the R software and further identified aging-related characterized genes by Lasso regression analysis.Construction and evaluation of the DNN model in the test dataset were then performed after model being optimized in the training dataset.Results A total of 243 DEG were confirmed. The age prediction model based on DNN by using 15 aging-related characterized genes showed the best prediction results with the maximum error of 11 years,the median error of 4 years,and the root mean square error of 6 years.Compared with ENR,RF,SGD,and SVR models,our model based on DNN showed the minimal error and the best precision.The 15 aging-related characterized genes were also confirmed in the aging HUVEC cell.Conclusion DNN model developed in this study can accurately and specifically predict age of healthy cases,which can play an important role in mechanism studies,also treatment,and prognosis in aging-related diseases.
万方数据
维普
