Identification Of Coding And Non-coding Sequences In A Complete Genome Using H(?)lder Exponent Formalism And Multiaffinity Analysis

Accurate prediction of genes in genomes has always been a challenging taskfor bioinformaticians and computational biologists. Therefore, the discovery ofexistence of distinct scaling relations in coding and non-coding sequences hasled to new perspectives in the understanding of the DNA sequences. This hasmotivated us to find new ways for characterization and classification of codingand non-coding sequences.In this thesis, we first introduce a number sequence representation of DNAsequences proposed by our group. Multiaffinity analysis and Holder formalism arethen performed on the representation of the obtained number sequence. Threesuited exponents are selected to form a parameter space. The two exponentsγ(-2),γ(6) are from Multiaffinity analysis, the exponent h is from Holder for-malism. Each coding or non-coding sequence may be represented by a point inthe three-dimensional parameter space. We can see the points corresponding tocoding and non-coding sequences in the complete genome of many prokaryotesbe divided to different regions roughly. If the point (γ(-2),γ(6), h) for a DNAsequence is situated in the region corresponding to coding sequences, the sequenceis discriminated as a coding sequence; otherwise, the sequence is classified as anon-coding one. Therefore these exponents can be used to distinguish coding andnon-coding sequences. The Fisher's discriminant algorithm is used to give thediscriminant accuracies. The average discriminant accuracies p_c, p_nc, q_c and q_nc ofall 51 prokaryotes obtained by the present method reach 66.53%, 83.34%, 71.63%and 83.54%. respectively.