Prediction of transcription factor binding sites in bacterial genomes: case study with Pectobacterium atrosepticum

The predicted hrp boxes include all expected hrp boxes upstream of well-characterized hrp genes, which gives confidence in the existing models to represent members of the hrp box family. Some candidate genes downstream of the putative hrp boxes are found to be downregulated in a HrpL mutant . Future work will investigate the combination of biological features into the models and generalize the predictor for all promoter binding sites on enterobacterial species. Conclusion Predicted HrpL-regulated genes Figure 2: hrp box MSA built using ClustalW [1]. • Building of multiple sequence alignment (MSA) using ClustalW (fig.2). • Generation of models with regular expressions (right) and HMM (using HMMER package [2]). The models are derived from the MSA (fig.2). • Model validation with 10-fold cross-validation method using test sets made up of 2000 random sequences, which were generated using either single nucleotide composition, or di-nucleotide composition of known promoter regions. A diagram of the architecture of the hrp box HMM profile is illustrated in figure 3. Modelling methods