Prediction of regulatory interactions in Arabidopsis using gene-expression data and support vector machines

Identification of regulatory relationships between transcription factors (TFs) and their targets is a central problem in post-genomic biology. In this paper, we apply an approach based on the support vector machine (SVM) and gene-expression data to predict the regulatory interactions in Arabidopsis. A set of 125 experimentally validated TF-target interactions and 750 negative regulatory gene pairs are collected as the training data. Their expression profiles data at 79 experimental conditions are fed to the SVM to perform the prediction. Through the jackknife cross-validation test, we find that the overall prediction accuracy of our approach achieves 88.68%. Our approach could help to widen the understanding of Arabidopsis gene regulatory scheme and may offer a cost-effective alternative to construct the gene regulatory network.

Research highlights
► Gene-expression data and SVMs are explored to predict regulatory interactions in Arabidopsis.
► Experimentally validated regulatory relationships were collected as the positive example.
► Negative training examples were randomly selected TF-target pairs under some strategies.
► Expression data of 79 experimental conditions were extracted as the sequence feature.
► Through the jackknife test, the overall accuracy of our prediction achieved 88.68%.