The Application of Feature weighting models for Identification of key genes associated with the Transcriptomic Response to Drought Stress in Populus Species

Introduction: Poplar varieties are planted in short rotation coppice, and are supposed to show high biomassproduction. Drought is a very important abiotic stressor. High throughput gene expression technologiesprovide valuable information about transcriptome. Feature weighting models are also known as attractivestrategies to gain new biological insights. At the transcriptome level, the algorithms for identifying keysignatures related to environmental stress have not been applied in Populus. In this study, we used the largetranscriptome data to gain comprehensive view of drought stress response in Populus.Method: The array expression datasets retrieved from GEO and ArrayExpress. RMA algorithm was used forbackground correction and normalization of gene expression data by Affy R package. Finally, an empiricalBayes method was performed to correct non-biological differences and remove batch effects from geneexpression datasets using ComBat function in the SVA Rpackage. Feature selection algorithms wereemployed to reduce the dimensionality of expression dataset and identify the gene expression features. Weimplemented various attribute weighting algorithms include SVM, Chi Squared, Information Gain,Information Gain Ratio, Deviation, Gini Index, Uncertainty, Relief, and PCA to identify the most importantgenes using RapidMiner Studio software.Result: In total ۱۳ microarray datasets consisting of ۳۲۴ arrays were considered. After pre-processing andremoving the batch effect, the normalized datasets were obtained for further downstream analysis. In total,۶۴۸ genes were identified as the most important features by at least one of the models. Functional annotationshowed that the feature genes were enriched in response to abiotic stimulus and MAPK signaling pathway.In addition, a lot of genes were related to secondary metabolic process. Interestingly, the seven methodsselected auxin response factor ۲-like and PYL۴-like as important features.Conclusion: Our analysis suggests that ARF۲-like and PYL۴-like genes can be potential candidates forscreening and breeding purposes in Populus.

Keywords:

Populus Species , Feature weighting models , SVM , PCA