Finding gene promoters in the genome of the fungus crinipellis perniciosa using feed-forward neural networks

The detection and structural characterization of genes in genome projects requires sophisticated automatic tools, most of then based on machine learning techniques. While functional genomics looks after the composition and function of the proteins codified by the genes, geneticists are more interested in investigating the mechanism, which regulates the expression of the genes. In particular, the study of the promoters is of crucial importance for understanding the responses to biological and environmental stimuli. In this article, we address the use of neural networks for promoter recognition in the genome of the fungus crinipellis perniciosa, an aggressive phytopathogen of the cacao tree, which is being sequenced by a Brazilian consortium. A divide and conquer strategy was used for the solution of the complex problem of localizing and characterizing the gene promoters. The division of the problem is based on the localization of an internal structure called TATA-box, considered as one of the promoter's signal. With that purpose, we trained a feed-forward neural network using patterns found in other species, due to absence of validated data for the fungus under study. A new approach for feature extraction, based on local compositional measures, is described. Currently, biological studies are being carried out for the experimental validation of the predictions of the neural network