It was lower for all methods when using the protein protein interaction category and higher when using the sequence cluster category (Supplementary Fig. S3A and Table S1).
The de novo transcripts from Trinity's output were compacted using the sequencing clustering program cd-hit (http://www.bioinformatics.org/cd-hit/) version 4.5.4 and option "'-c 0.95 -n 9 -g 1 -M 0 -T 20".
To deal with amino acid sequences and BLOSUM matrices, a trivial approach would be to use the sequence clustering rule defined by Henikoff and Henikoff [ 39]: to compute the BLOSUM η similarity matrix, they grouped amino acid sequences into clusters in such a way that each sequence in any cluster has at least η% identity to at least one other sequence in this cluster.
Local clustering of the ESTs was performed using the sequence-based clustering program UIcluster [ 41], allowing matches based on both the forward and reverse complements.
Users can associate new sequences to a pre-aligned set of clusters using the sequence search option.
Sequences were clustered using the sequence assembly program CAP3 (Huang and Madan, 1999) with default parameters, into a set of 128 nonredundant contigs and 401 singletons (Appendix S1 and S2).
In addition, the presence of genomic clusters in the D. solani and D. dadantii species was searched with a nucleotide BLAST analysis using the sequences of the specific clusters as queries against a database constituted by the D. solani and D. dadantii genomes (e-value threshold = 10-50).
The genomic locations of the sequence clusters were used to determine the distribution of the coding regions in the genome.
The sequences encoding a putative transposase identified using these profiles are extracted and clustered using the UCLUST [ 17] softwares and the sequence clusters are finally aligned with MUSCLE [ 20].
To determine the quality of our classification we used the sequence comparisons from the clustering to find the median identity for the best hit among the classified and the unclassified proteins, respectively.
