Computer Science Honours Research Report Compression and Computational Gene Finding

Gene sequences in DNA are punctuated with regions of “junk” that are not used during expression of the gene. Identifying these regions is a complex task and many computational techniques have been devised to solve the gene finding problem. DNA can be described as a sequence over an alphabet of four letters {A, G,C, T} and so a sequence can be considered as a text written in some language. Gene finding methods that use this linguistic approach have been very successful. This research investigated the feasibility of a linguistic approach based on compression. Compression algorithms have been successfully used in the linguistic analysis of human texts, including the differentiation between texts based on language and author. The useful and junk parts of a DNA sequence can be substantially different and can be viewed as two different languages. The research used a compression-based measure of entropy to attempt to differentiate between coding and non-coding regions in DNA sequences. Additionally, the same measure was used to compare genes from different families and parts of different species’ genomes. The results show that the measure is unable to identify the often subtle variations in genomic data, preventing it from effectively discriminating between different types of DNA sequences. It is reasoned that the measure is not suited to genomic data in general and has limited applications in the field of bioinformatics. Possible reasons for this include the small alphabet and correspondingly limited range of “words” in the sequences. This report details the research performed and discusses the implications of the results.