Five of the selected families were excluded, either because they had expanded in a timeframe not interesting for this study or had a multitude of members and/or domains that made alignments unreliable.
The other paralogs were omitted because they were either missing a domain or have large insertions that make alignments unreliable.
Before we aligned the avian ZPC introns we removed regions of simple repeats, identified by the program Sputnik [ 25] since they make alignments unreliable and are a potential cause of bias due to elevated substitution and indel rates.
However, automated sequence alignment is unreliable at high divergence levels, so we excluded orthologous pairs with less than 30% identity (leaving 4,975 contigs) to avoid false mismatches introduced by low confidence alignments.
Margulies et al. [56] argued that pairwise alignments are unreliable for detecting regulatory elements.
The remaining sequences were aligned with MAFFT [39] and unreliable alignment positions were removed with GUIDANCE [40].
In spite of this, because our aligned sequences are quite conserved, it is unlikely that an unreliable alignment might have caused false positive identification of sites potentially under selection [ 30].
The detected unreliable alignment regions were discarded.
The amino acid data set was manually inspected to isolate areas of unreliable alignment.
To avoid these potential errors, Gblocks (Castresana 2000) was used to detect and filter unreliable alignment regions.
