Abstract

We present the concepts and design outline used in the development of an advanced system for promoter recognition of vertebrates. The system uses a collection of models. Each of the models is based on the multi-sensor integration and an artificial neural network (ANN). The models extract relevant information for the assessment of promoter presence in the query sequence. The new sensors developed are based on the statistical concept of oligonucleotide distributions in specific functional regions of DNA. These distributions are modeled as a set of positional weight matrices of the most significant oligonucleotides. A query sequence is analyzed by sliding a data-window along it, and presenting segments of data (contained in the data-windows) to the system for the assessment of promoter presence. This data is then passed through the system sensors and their outputs are presented to the ANN for the final processing. Based on the level of the ANN output signal, the data-window is classified as belonging to the promoter or not. The system is evaluated on a very large and diverse human sequence-set and exhibits several times higher accuracy than the other publicly available promoter recognition systems. The system is implemented in the Dragon Promoter Finder package, which can be accessed at http://sdmc.krdl.org.sg:8080/promoter/ Keywords: DNA analysis, promoter recognition, gene finding, multi-sensor integration, artificial neural networks, composite models. 1 To whom correspondence should be directed V B Bajic et al. Intelligent system for promoter recognition 2 Notation. bp -- base-pair Mbp -- one million bp nt - nucleotide TP -- true positive FP -- false positive DNA -- deoxyribonucleic acid RNA -- ribonucleic acid TSS -- transcription start site TF -- transc...

Citations