9 Genomic Techniques and How to Apply Them to Marine Questions
333
gene identification, GISMO employs a Support Vector Machine, which is able to
learn sequence properties of different classes of genes in an unsupervised manner.
Several programs have been devised for the prediction of translation start sites,
including GS-Finder (Ou et al. 2004), RBS-Finder (Suzek et al. 2001), and Tico
(Tech and Meinicke 2006), which have all achieved accuracy values of more than
90%. However, owing to the limited number of experimentally confirmed translation initiation sites that were available for a performance evaluation, these accuracy
values cannot be generalized.
The online resources REGANOR (Linke et al. 2006) and RAST (Aziz et al. 2008)
provide easy means to automatically predict tRNA, rRNA and protein encoding
genes in prokaryotic genomes via a web-interface. REGANOR achieves sensitivity values of up to 98% for “certain” genes with known gene function and a
specificity of 95% (McHardy et al. 2004a) by combining the gene finders Glimmer2 (Delcher et al. 1999) and CRITICA (Badger and Olsen 1999). The RAST
server on the other hand combines evidence from different sources to identify
protein-encoding genes, including Glimmer-2 predictions, homology searches for
conserved protein families (FIGfams), and a BLAST search versus a large protein database. Additionally, RAST provides an automated functional annotation
of all genes that are detected. Both the REGANOR and RAST server employ
the programs tRNAscan-SE (Lowe and Eddy 1997) and SearchForRNAs (Niels
Larsen et al., unpublished), which allow the automated identification of tRNA and
rRNA genes in raw genomic sequences. The REGANOR server can be accessed at
https://www.cebitec.uni-bielefeld.de/groups/brf/software/reganor/, the RAST server
at http://rast.nmpdr.org/.
In summary, the gene finding web-servers REGANOR and RAST provide high
quality gene predictions and are easy to use. The public gene finders GLIMMER
and GISMO on the other hand can be installed locally, enabling their integration
into existing genome annotation pipelines. While Glimmer is easy to use and timeeffective, GISMO is slower and more difficult to maintain, but provides highly
accurate gene calls. tRNAscan-SE (Lowe and Eddy 1997), SearchForRNAs and
RNAmmer (Lagesen et al. 2007) allow the automated identification of tRNA and
rRNA genes (Table 9.3).
9.3.2.2 Gene Finding in Eukaryotes
Although improvements in eukaryotic gene finding programs have been made during recent years, prediction of gene structure in eukaryotic genomes is still a highly
difficult task. The main challenges are the complex structure of eukaryotic genes and
the low fraction of chromosomes corresponding to protein encoding exons (Mathe
et al. 2002, Zhang 2002, Brent 2007). While about 90% of prokaryotic genomes
encode proteins, eukaryotic exons are embedded in vast amounts of non-coding
DNA. The task is further complicated by the fact that eukaryotic genes may have
alternative splice and polyadenylation sites, as well as alternative transcription and
translation initiation sites.
333
gene identification, GISMO employs a Support Vector Machine, which is able to
learn sequence properties of different classes of genes in an unsupervised manner.
Several programs have been devised for the prediction of translation start sites,
including GS-Finder (Ou et al. 2004), RBS-Finder (Suzek et al. 2001), and Tico
(Tech and Meinicke 2006), which have all achieved accuracy values of more than
90%. However, owing to the limited number of experimentally confirmed translation initiation sites that were available for a performance evaluation, these accuracy
values cannot be generalized.
The online resources REGANOR (Linke et al. 2006) and RAST (Aziz et al. 2008)
provide easy means to automatically predict tRNA, rRNA and protein encoding
genes in prokaryotic genomes via a web-interface. REGANOR achieves sensitivity values of up to 98% for “certain” genes with known gene function and a
specificity of 95% (McHardy et al. 2004a) by combining the gene finders Glimmer2 (Delcher et al. 1999) and CRITICA (Badger and Olsen 1999). The RAST
server on the other hand combines evidence from different sources to identify
protein-encoding genes, including Glimmer-2 predictions, homology searches for
conserved protein families (FIGfams), and a BLAST search versus a large protein database. Additionally, RAST provides an automated functional annotation
of all genes that are detected. Both the REGANOR and RAST server employ
the programs tRNAscan-SE (Lowe and Eddy 1997) and SearchForRNAs (Niels
Larsen et al., unpublished), which allow the automated identification of tRNA and
rRNA genes in raw genomic sequences. The REGANOR server can be accessed at
https://www.cebitec.uni-bielefeld.de/groups/brf/software/reganor/, the RAST server
at http://rast.nmpdr.org/.
In summary, the gene finding web-servers REGANOR and RAST provide high
quality gene predictions and are easy to use. The public gene finders GLIMMER
and GISMO on the other hand can be installed locally, enabling their integration
into existing genome annotation pipelines. While Glimmer is easy to use and timeeffective, GISMO is slower and more difficult to maintain, but provides highly
accurate gene calls. tRNAscan-SE (Lowe and Eddy 1997), SearchForRNAs and
RNAmmer (Lagesen et al. 2007) allow the automated identification of tRNA and
rRNA genes (Table 9.3).
9.3.2.2 Gene Finding in Eukaryotes
Although improvements in eukaryotic gene finding programs have been made during recent years, prediction of gene structure in eukaryotic genomes is still a highly
difficult task. The main challenges are the complex structure of eukaryotic genes and
the low fraction of chromosomes corresponding to protein encoding exons (Mathe
et al. 2002, Zhang 2002, Brent 2007). While about 90% of prokaryotic genomes
encode proteins, eukaryotic exons are embedded in vast amounts of non-coding
DNA. The task is further complicated by the fact that eukaryotic genes may have
alternative splice and polyadenylation sites, as well as alternative transcription and
translation initiation sites.