ORFs as Markov chains

Assuming we found all ORFs in a sequence, we can use codon frequencies to find which ORFs are coding and which are non coding open reading frames (NORFs). We translate each ORF into a codon sequence and get 64-state Markov chain. We use a state for each codon rather than a state for each amino acid, because codons are more informative than their translations. (There might be a preference for a specific codon in gene expression over other codons that encode the same amino acid ). The transition probabilities are the probabilities for each codon to follow any other codon in a coding region. Using this model, we can compute the probability that a given ORF is really a coding region. Figure shows the results of such method.

  

Figure: Coding regions recognition using 64 state markov model [].