In living organisms, gene expression is finely regulated by the joint action of regulatory proteins. Among these proteins, transcription factors (TFs hereafter) play a key role since they bind specific sequences in the promoters of genes to initiate their regulation. TFs can combine to form complexes and regulate the expression of new genes or to alter the direction or level of regulation of genes already targeted by one of the two TFs alone. The complexes thus diversify the repertoire and regulatory levels of genes targeted by the transcription factors. Little is known about the extent of this phenomenon, the number of complexes, the identity of the partners and the way they bind to DNA. This project proposes to develop a bioinformatics model to predict the existence of protein complexes formed by transcription factors and likely to regulate gene expression in the plant Arabidopsis thaliana. In a second phase, the project will explore the predictions of the model to verify the existence of the predicted complexes, and to characterize their DNA binding mode and their target genes. The discovery of new complexes will be done by developing a model that integrates clues scattered in different types of genomic data. These clues are (i) the common binding of the TFs on promoter regions, the motifs and combinations of DNA motifs bound by the TFs on these bound regions, (i) the co-expression of the TFs, (iii) the target genes common to both TFs, and (iv) the co-evolution of amino acid residues between the two TFs forming a complex. The model will be obtained by machine learning on these data: the model will be built and its parameters adjusted to optimize the predictions against a set of TFs known to form complexes. Newly predicted interactions, in particular those between transcription factors studied in our lab and new partners, will be explored in detail to understand how these complexes form (interaction surface), how they bind DNA and to know which genes and functions they regulate. The results of the model will be represented in the form of an interaction network for all Arabidopsis thaliana TFs.. This network will be made available to the community so that biologists can in turn explore the potential partners of their favorite TFs. In the medium term, the model could be applied to other plant species such as rice and maize, two species characterized by extensive genomic data. This approach represents a considerable time saving compared to the genetic method and works even in the case where several TFs play a redundant role.