Prediction the Age of Human Brains from Gene Expression

Abstract

Understanding temporal characteristics of gene expression in normal human brain can help explain the neurodevelopment, working mechanism and functional diversity. Based on the gene expression dataset of developing human brains from the Allen Brain Atlas, we accurately predicted the age of human brains using support vector machine and identified 9,934 age related genes. Significant changes occur in gene expression of human brains before and after birth, thus we establish support vector machine (SVM) models for the subjects before birth and after birth, respectively. In general, the age of subjects can be well predicted by the SVM models, with the Pearson correlation coefficient of predicted age and the labeled age of all subjects is 0.9397 with P-value < 0.001 (before birth: r = 0.9465, P-value < 0.001; after birth: r = 0.9121, P-value < 0.001). For the total subjects, mean absolute error (MAE) of age prediction is 2.82 years with standard error (SE) is 0.15 years (before birth: MAE = 1.03 post-conceptual weeks (pcws), SE = 0.08 pcws; after birth: MAE = 4.70 years, SE = 0.20 years). This investigation reveal the bulk of temporal regulation occurred during prenatal development. By analyzing the functional annotations of age related genes, we found expression differences of genes before and after birth may be related to their functions. Finally, we found the prediction accuracy of each period can reflect its specificity of gene expression, which is negatively correlated to the gene expression similarity across periods. This study provides new insights into temporal dynamic pattern of gene expression in human brains and its relationship with functions.