Evaluation of First and Second Markov Chains Sensitivity and Specificity as Statistical Approach for Prediction of Sequences of Genes in Virus Double Strand DNA Genomes

Abstract

Growing amount of information on biological sequences has made application of statistical approaches necessary for modeling and estimation of their functions. In this paper, sensitivity and specificity of the first and second Markov chains for prediction of genes was evaluated using the complete double stranded DNA virus. There were two approaches for prediction of each Markov Model parameter, initial probability and transition matrix, which together with the first and second Markov chains resulted in development of eight algorithms for gene prediction. In order to compare the algorithms, a sensitivity and specificity repeated measure with 3 factors (Markov model, type of selection and estimation of transition probabilities) were utilized. Results significantly revealed that the second order Markov chain had more sensitivity and specificity than the first order Markov chain, with “p-Value" < 0.001. By adding the covariates, the number of annotated genes per length of genome as well as the A & T and C & G contents of genomes in the repeated measure showed an insignificant difference between the sensitivities of the two Markov models (0.407, 0.071 and 0.120, respectively). It was also proved that gene base-pairs per genome length and A & T contents of the genome, as model covariates, resulted in significant differences between the specificities of the Markov models.