We analyzed the conservation of TGA-flanking regions in all six reading frames using BLASTX. If the best hit, which covered the TGA codon with at least a 10-nt overlap, was in a different reading frame than the TGA codon, the corresponding sequence was filtered out. RPS-BLAST was then used to search against conserved domains database (CDD). If the best hit which covered the TGA codon with at least a five-residue overlap was in a different reading frame or additional stop codons appeared within the conserved domain in the same frame, the sequence was removed.