To select a non-redundant set, we used mkvtree in the Vmatch suite (see URLs) to index the sequences by length; Vmatch was used to cluster sequences, including clusters with size of 1 (singlets).
We additionally clustered sequences by length to see if the sheer number of insertions and deletions in divergent sequences were mainly responsible for this effect.
A further quality check was conducted for the final set of putative transcripts by filtering sequences by length (minimum 200 bp) and average sequence quality (minimum of 30).
Most methods introduced herein, including programs in CD-HIT package, Uclust and DNACLUST sort sequences by length and process them from long to short.
This program creates clusters using a greedy algorithm, first selecting seed sequences by length, then finding sequences with identities to a particular seed at greater than or equal to a specified threshold value (for example 60%).
The density of repetitive sequences (calculated by dividing the total length of repetitive sequences by the length of the clusters analyzed) in the caecilian Hox clusters was 3.3 %, among the lowest densities of all the species studied (Fig. 2).
Nucleotide sequences were translated into amino acid sequences, and filtered by length, selecting only sequences >400 amino acids.
In the particular case where several sequences were present in the same species, sequences were ordered by length.
CD-HIT implements a greedy incremental clustering algorithm, which orders the sequences by decreasing length and makes the longest sequence the representative sequence of the first cluster.
We then removed redundancy within the MGC clone set, and picked the clone closest to the longest reference sequence by CDS length as a best MGC representative for each gene.
To meet this requirement, we preprocess the database by sorting the subject sequences by the length in the ascending order.
