Data sets
To evaluate our multidimensional scaling (MDS) approach, we focused on visualizations of ribosomal protein genes and putative alien genes for different microbial genomes. Ribosomal protein genes belong to the class of highly expressed genes which tend to use codons associated with the prevalent tRNAs present in the organism. If translational selection is one of the main sources for codon preferences in a particular genome, then codon usage can in turn be used for the prediction of putative highly expressed genes [20]. Another source of codon usage variation in microbial genomes is provided by the insertion of foreign DNA by means of horizontal gene transfer. Thus, putative alien genes may also be predicted on the basis of codon usage analysis [2,21]. While ribosomal protein genes can be identified from the annotations of completely sequenced genomes, reliable information about putative alien genes is much more difficult to obtain. We combined predictions of the SIGI-HMM tool [22] with existing references from the literature in order to obtain suitable test sets for our evaluations. SIGI-HMM is based on a Hidden Markov Model for the detection of genomic islands and, in contrast to our MDS-based visualization method, it explicitly uses information about the locations of genes on the corresponding chromosomes. However, unlike MDS, SIGI-HMM does not consider codon usage correlations between different amino acids. Using the two complementary kinds of information exclusively, both methods provide completely different approaches to codon usage analysis [22].
For the evaluation of the MDS-based visualizations we analyzed the microbial genomes of Escherichia coli K-12, Bacillus subtilis, Vibrio cholerae and Thermus thermophilus HB8. We used annotated DNA sequence data in the EMBL format publicly available from EBI [23]. Ribosomal protein genes were extracted from the datasets of the completely annotated genomes. Putative alien genes were selected according to the following information: On chromosome 1 of V. cholerae two genomic islands were predicted by SIGI-HMM that comprise a gene cluster for a toxin-coregulated pilus and fragments of a temperate filamentous phage described in [24]. Both clusters are closely associated with the pathogenicity of V. cholerae. For Bacillus subtilis 10 integrated prophages have been described based on experimental evidence and theoretical considerations [25-28]. Nine of these prophages overlap with genomic islands as predicted by SIGI-HMM. For Escherichia coli K-12 the authors of [29] used different compositional variables and estimated that about 18% of the genome have been imported by horizontal gene transfer. In contrast, SIGI-HMM predicted 580 genes (13,6%) to be putatively alien. The largest genomic islands comprise the cryptic prophages CP4-6, DLP12, e14, Rac, Qin, CP4-44, CPS-53, Eut, CP4-57, and the phage-like element KpLE2 (reviewed in [30]). For the extremophilic bacterium Thermus thermophilus HB8 no genomic islands have been described so far. SIGI-HMM predicted a contiguous gene cluster of 5 genes associated with functions in cell wall biosynthesis to be putative alien. The total number of putative alien genes and the number of ribosomal protein genes for all species considered here are summarized in table 1. Additional file 1 provides a detailed list of all putative alien genes used for the visualization.
Table 1  Number of genes used for the visualization for all species under consideration. Given are the number of putative alien genes, the number of ribosomal protein genes and the total number of genes on the respective chromosomes.
species  # genes (total)  # ribosomal protein genes  # putative alien genes
E. coli  4254  61  206
B. subtilis  4106  57  317
V. cholerae Chr1  2736  64  41
V. cholerae Chr2  1092  0  216
T. thermophilus  1973  60  5