Using AI techniques to determine promoter location based on DNA structure calculations

Abstract

DNA sequencing projects have started the race to fully annotate complete genomes, including the human one. Despite that, little is known about genetic regulation, the mechanisms that control where and when the genes are expressed, and promoters are maybe the most important of these mechanisms. An increasing number of studies have been focused on the DNA molecule and its structure. This has lead to a set of physical properties which can be computed from mathematical models, and describe some aspects of this molecule. Unfortunately, the existing tools are scattered through the different web sites of many research groups, and extracting data with them is still very unpleasant. The first part of this thesis presents DNAlive, a new platform to calculate DNA physical properties, showing the results in a visual and useful way for genetic researchers, cross-linking the data with external databases. For the second part, a full study of DNA physical descriptors has been performed, revealing significative similarities between them. Using that data, a set of neural networks has been developed to detect promoters on a DNA sequence. The resulting software is the second version of ProStar, the MMB group's1 latest promoter predictor.