Professor Klaas Vandepoele presents a novel supervised learning approach that predicts transcription factor functions

Plant development and plant growth rely on the careful integration of internal signals and external environmental information. In response to these cues, gene expression is tightly controlled by gene-regulatory networks (GRNs), which entail transcription factors (TFs). To date, many different experimental approaches and -omics studies have been applied to understand the interactions between TFs and their target genes, which together regulate various developmental processes, stress responses and growth. In a novel approach, research led by Professor Klaas Vandepoele makes use of machine learning to build integrated GRNs (iGRNs) to predict and understand the functional relationships between TFs and their target genes. In the first outcome of this supervised learning method, almost each single gene encoded in the flowering plant A. thaliana could be given a systematic functional and regulatory annotation in an iGRN.

Professor Vandepoele visited the Oxford Plant Sciences Department to present research that he is conducting at the University of Gent. Using publicly available experimental and -omics input networks, he computes iGRNs and trains them to predict TF target genes as well as associated TF functions. Strikingly, these iGRNs outperform all input networks in predicting known regulatory interactions. This computational approach was developed and verified in A. thaliana, where it provides high sensitivity and specificity in predicting TF functions. Now, Professor Vandepoele and his research team aim to also build iGRNs in other plant species ranging from liverworts to crop plants. The biggest challenge in this undertaking is the limited availability of input data. However, if successful, different species-specific iGRNs will enable us to more effectively and comprehensively study the properties and evolution of GRNs in plants. Professor Vandepoele demonstrated that feeding experimental biological data into a machine learning algorithm can provide a fast and complex understanding of gene regulation. Since experimental data on their own have not been able to achieve that, his new approach promises great advances towards understanding functional regulatory interactions within GRNs.

List of site pages