Figure 1. Examples of complexity charts used for DNA sequence segmenting and approximate functional annotation. Both charts were generated with a window width, W, of 200 nucleotides moving one nucleotide at a time (window step, s=1). The accuracy of correctly annotated positions is too low (100 nucleotides) to be useful for exact gene structure determination, but it is clear that compositional complexity is correlated with gene structure. (a) Modified compositional complexity chart (z‐score of MCC) for the region analogous to the α‐operon of Escherichia coli in halophilic archaea, Halobacterium halobium. Arrows with transparent points indicate probable intergenic regions between protein‐coding sequences. (b) Local compositional complexity and modified compositional complexity in chicken ovalbumin gene X. Arrows with filled points indicate probable positions of introns. Arrows with transparent points show the false‐positive indications of intergenic spacers. (Determining the number of different genes in putative protein‐coding regions is a serious problem that plagues all computer‐assisted gene prediction methods. Compositional complexity chart methods face this problem as well.)

Figure 2. Slopes of straight line fits for surprisal versus complexity data for short oligonucleotides (regular and patchy) in large samples of human exons of confirmed protein‐coding genes, introns, 3′ untranslated regions (UTRs) and 5′ UTRs of these genes. The x coordinate of each plot in the ‘matrix’ represents patchiness (k=0, 1,…,9). The y coordinate represents slope values of surprisal versus complexity regression line. (All slope values are significant at a confidence level of 5% or better.) In every figure panel, block lengths L=5, 9, 13 and 20 correspond to the top to bottom lines respectively. Figure panels in the ‘exons’ column of the matrix show that exons display clear three‐base periodicity of occurrence of short oligonucleotides at all levels of patchiness, in all four alphabets. Comparison of ‘introns’ and ‘3′ UTRs’ columns shows that the complexity‐related properties of introns and 3′ UTRs are remarkably similar in most cases. This explains known difficulties with determining number of protein‐coding genes in computationally predicted ‘coding regions’. The only significant differences (and precious for practical purposes of gene identification) can be found for 20‐grams in {A, C, G, T}, {K, M} and {R, Y} alphabets. Comparison of ‘exons’ and ‘5′ UTRs’ columns also shows that complexity‐related properties of exons and 5′ UTRs are similar enough to cause problems with computational identification of 5′ ends of protein‐coding genes. Comparison of figure panels in the bottom right and the bottom left corners of the matrix suggests that using statistics of 20‐grams in the {S, W} alphabet should help to correctly identify 5′‐UTRs correctly.