AUTHOR=Chai Keping , Liang Jiawei , Zhang Xiaolin , Cao Panlong , Chen Shufang , Gu Huaqian , Ye Weiping , Liu Rong , Hu Wenjun , Peng Caixia , Liu Gang Logan , Shen Daojiang TITLE=Application of Machine Learning and Weighted Gene Co-expression Network Algorithm to Explore the Hub Genes in the Aging Brain JOURNAL=Frontiers in Aging Neuroscience VOLUME=13 YEAR=2021 URL=https://www.frontiersin.org/journals/aging-neuroscience/articles/10.3389/fnagi.2021.707165 DOI=10.3389/fnagi.2021.707165 ISSN=1663-4365 ABSTRACT=

Aging is a major risk factor contributing to neurodegeneration and dementia. However, it remains unclarified how aging promotes these diseases. Here, we use machine learning and weighted gene co-expression network (WGCNA) to explore the relationship between aging and gene expression in the human frontal cortex and reveal potential biomarkers and therapeutic targets of neurodegeneration and dementia related to aging. The transcriptional profiling data of the human frontal cortex from individuals ranging from 26 to 106 years old was obtained from the GEO database in NCBI. Self-Organizing Feature Map (SOM) was conducted to find the clusters in which gene expressions downregulate with aging. For WGCNA analysis, first, co-expressed genes were clustered into different modules, and modules of interest were identified through calculating the correlation coefficient between the module and phenotypic trait (age). Next, the overlapping genes between differentially expressed genes (DEG, between young and aged group) and genes in the module of interest were discovered. Random Forest classifier was performed to obtain the most significant genes in the overlapping genes. The disclosed significant genes were further identified through network analysis. Through WGCNA analysis, the greenyellow module is found to be highly negatively correlated with age, and functions mainly in long-term potentiation and calcium signaling pathways. Through step-by-step filtering of the module genes by overlapping with downregulated DEGs in aged group and Random Forest classifier analysis, we found that MAPT, KLHDC3, RAP2A, RAP2B, ELAVL2, and SYN1 were co-expressed and highly correlated with aging.