Finding long ORFs

One way to distinguish coding regions from non-coding regions, is to examine the frequencies of stop codons. Assuming a uniform random distribution, a stop codon is expected to be observed every $64/3\approx 21$ codons (since there are 3 stop codons). Average proteins are much longer, being coded by about 1000bp (base pairs). Each coding region has only one stop codon, which terminates the region. Therefore, one way to detect the coding regions, is to look for long sequences of codons, without any stop codon. The algorithm that uses the above idea scans the DNA sequence, looking for long ORFs in all three reading frames. Upon detecting a stop codon, the algorithm scans backward, searching for a start codon. This algorithm will fail to detect very short genes, as well as overlapping long ORFs on opposite strands. Moreover, there are a lot more ORFs than genes. For example, we can find 6500 ORFs in the DNA of the bacterium E.coli while there are only 4400 genes.