Neural-statistical Model of Tata-box Motifs in Eukaryotes

The TATA-box is one of the most important binding sites in eukaryotic Polymerase II promoters. It is also one of the most common motifs in these promoters. The TATA-box is responsible mainly for the proper localization of the transcription start site (TSS) by the biochemical mechanism of DNA transcription. It also has very regular distances from the TSS. Accurate computational recognition of the TATA-box can improve the accuracy of the determination of the TSS location by computer algorithms. The conventional recognition model of the TATAbox in DNA sequence analysis is based on the use of a position weight matrix (PWM). The PWM model of the TATA-box is widely used in promoter recognition programs. This chapter presents a different, nonlinear, recognition model of this motif, based on a combination of statistical and neural network modelling. The resulting TATA-box model uses “statistical filtering” and two LVQ neural networks. The model is derived for a sensitivity level that corresponds to approximately correct recognition of TATA motifs. The system is tested on an independent data set used in the evaluation study by Fickett and Hatzigeorgiou, and it performs better in promoter recognition than three other methods, including the one based on the matching score of the TATA-box PWM of Bucher.