The good people at SILVA have released a new version of the SILVA database. A little bit of tweaking is needed to get their files to be compatible with mothur. This README document describes the process that I used to generate the mothur-compatible reference files.
Curation of references
Getting the data in and out of the ARB database
This README file explains how we generated the silva reference files for use with mothur’s classify.seqs and align.seqs commands. I’ll assume that you have a functioning copy of arb installed on your comptuer. For this README we are using version 6.0. First we need to download the database and decompress it. From the command line we do the following:
This will launch us into the arb environment with the ‘‘Ref NR 99’’ database opened. This database has 534,968 sequences within it that are not more than 99% similar to each other. The release notes for this database as well as the idea behind the non-redundant database are available from the silva website. Within arb do the following:
- Click the search button
- Set the first search field to ‘ARB_color’ and set it to 1. Click on the equal sign until it indicates not equal (this removes low quality reads and chimeras)
- Click ‘Search’. This yielded 472365 hits
- Click the “Mark Listed Unmark Rest” button
- Close the “Search and Query” box
- Now click on File->export->export to external format
- In this box the
Exportoption should be set to
Compressionshould be set to
- In the field for
Choose an output file name entermake sure the path has you in the
arb_ref_119folder and enter
Select a format: fasta_mothur.eft. This is a custom formatting file that I have created that includes the sequences accession number and it’s taxonomy across the top line. To create one for you will need to create
/opt/local/share/arb/lib/export/folder with the following:
SUFFIX fasta BEGIN
- You can now quit arb.
Screening the sequences
Now we need to screen the sequences for those that span the 27f and 1492r primer region, have 5 or fewer ambiguous base calls, and that are unique. We’ll also extract the taxonomic information from the header line:
The mothur command above does several things. First the screen.seqs command removes sequences that are not full length and have more than 5 ambiguous base calls. Note: this will remove a number of Archaea since the ARB RN reference database lets in shorter (>900 bp) archaeal 16S rRNA gene sequences. Second, pcr.seqs convert any base calls that occur before position 1044 and after 43116 to
. to make them only span the region between the 27f and 1492r priming sites. Finally, it is possible that weird things happen in the alignments and so we unalign the sequences (degap.seqs) and identify the unique sequences (unique.seqs). We then convert the resulting fasta file into an accnos file so that we can go back into mothur and pull out the unique sequences from the aligned file (get.seqs).
Formatting the taxonomy files
Now we want to make sure the taxonomy file is properly formatted for use with mothur.
Building the SEED references
The first thing to note is that SILVA does not release their SEED; it is private. By screening through the ARB databases we can attempat to recreate it. Our previuos publications show that classify.seqs with the recreated SEED does an excellent job of realigning sequences to look like they would if you used SINA and the true SEED. Now we want to try to figure out which sequences are part of the seed. Earlier, when we exported the sequences from ARB, we included the align_ident_slv field from the database in our output. Let’s generate an accnos file that contains the names of the seuqences with 100% to the SEED database and then use mothur to generate SEED fasta and taxonomy files. While we’re at it we’ll also generate the nr_119 taxonomy file as well…:
Let’s look to see how many different taxa we have for each taxonomic level within the
We see that our full-length database retains a signigicant majority of the taxa that were in the original NR database. The Archaea take a beating. If you are interested in analyzing the Eukaryota, I would suggest duplicating my efforts here but modify the screen.seqs and pcr.seqs steps to target your region of interest.
Finally, we want to compress the resulting alignment and this README file into the full length and SEED archives:
So… which to use for what application? If you have the RAM, I’d suggest using
align.seqs. It took about 10 minutes to read in the database file and a minute or so to align a 1000 full-length sequences. Here is an example workflow for use within mothur that will get you the V4 region of the 16S rRNA gene:
This will get you 104,711 unique sequences to then align against (meh.). Other tricks to consider would be to use
get.lineage to pull out the reference sequences that are from the Bacteria, this will probably only reduce the size of the database by ~10%. You could also try using
filter.seqs with vertical=T; however, that might be problematic if there are insertions in your sequences (can’t know a priori). It’s likely that you can just use the
silva.seed_v119.align reference for aligning. For classifying sequences, I would strongly recommend using the
silva.nr_v119.tax references after running pcr.seqs on
silva.nr_v119.align. I probably wouldn’t advise using
unique.seqs on the output.
[ Update November 19, 2014: Updated the Building the SEED references section to respond to Elmar’s comments ]