Rapid identification of taxonomic assignments (RITA)
RITA is a standalone software package and Web server for taxonomic assignment of metagenomic sequence reads. By combining homology predictions from BLAST or UBLAST with compositional classifications from a Naive Bayes classifier, RITA is able to achieve very high accuracy on short reads. Unlike other hybrid approaches which combine these predictions for all sequences to be classified, RITA uses a pipeline to first identify cases where both types of classifier are in agreement, which constitute the highest-confidence set. Sequences not classified in this manner are subjected to a series of downstream classification steps.
This work has been accepted for publication:
MacDonald NJ, Parks DH, and Beiko RG. Rapid identification of taxonomic assignments. Accepted to 'Nucleic Acids Research' April 4, 2012.
If you have any questions or bug reports, please let us know at <firstname.lastname@example.org>.
For smaller datasets, taxonomic attributions can be obtained with the RITA web server.
RITA is released under the Creative Commons Share-Alike Attribution 3.0 License.
- RITA v1.0.1 RITA source code
Prerequisites: You must have BLAST+ 2.2.21 or higher installed.
- These instruction assume the following directory structure:
/home/<user>/RITA/ FastTree/ FCP/ mothur/ rita/
- Unzip RITA:
> unzip RITA_v1_0_1.zip
- Download, unzip, and run FCP install (FCP_install.py) with '--protein ncbi_genomes.faa' flag:
> unzip FCP_1_0_3.zip -d ./FCP > cd FCP > python FCP_install.py --protein ncbi_genomes.faa
- Concatenate the files in the training/sequences folder of FCP into a single input nucleotide file (e.g. ncbi_genomes.fna):
> cat ./training/sequences/*.fasta > ncbi_genomes.fna
- RITA is designed to run over multiple BLAST databases in order to reduce memory consumption. Therefore, you should split both the nucleotide and protein files into X pieces using the splitfasta.py script in the
scripts directory (X can be 1 if you do not wish to split the files, we recommend X=10):
> cd ../rita > mv ../FCP/ncbi_genomes.faa . > mv ../FCP/ncbi_genomes.fna . > cd scripts > python splitfasta.py ../ncbi_genomes.fna 10 > python splitfasta.py ../ncbi_genomes.faa 10
- Create a nucleotide database with makeblastdb BLAST+ for each input ncbi_genomes.p*.fna:
> makeblastdb -in "ncbi_genomes.p1.fna" -dbtype nucl > makeblastdb -in "ncbi_genomes.p2.fna" -dbtype nucl > ... > makeblastdb -in "ncbi_genomes.pX.fna" -dbtype nucl
- Create a protein database with makeblastdb BLAST+ for each input ncbi_genomes.p*.faa.
> makeblastdb -in "ncbi_genomes.p1.faa" -dbtype prot > makeblastdb -in "ncbi_genomes.p2.faa" -dbtype prot > ... > makeblastdb -in "ncbi_genomes.pX.faa" -dbtype prot
- Set BLASTDB_PARTS to X in globalconfig.cfg and the name of the database to
<database_name>.p%%d (e.g. ncbi_genomes.p%%d.fna if this was your output BLAST database name) %%d is the placeholder for the database number identifier, 1..X.
- Configure globalsettings.cfg appropriately for the above installation directories.
- To use the UBLASTX classifier, you must also obtain a licensed copy of usearch and set up the configuration file appropriately.
Rank-specific example usage
python rita.py --rank PHYLUM --pipeline NB_DCMEGABLAST,DCMEGABLAST_RATIO,NB_RATIO,NB_ML --query fragments.fasta --out results.txt
Note: Use the --jobid flag with a unique job identifier if running rita in parallel to ensure intermediate temporary files are not overwritten.
The above command will classify the fragments in 'fragments.fasta' at the rank of PHYLUM, using a pipeline that starts with the consensus of Naive Bayes and DCMEGABLAST, then the most confident DCMEGABLAST results, then the most confident Naive Bayes results and finally with the maximum likelihood NB prediction. Note that fragments are not attempted to be classified at a given step in the pipeline if they have already been classified at an earlier step (the order matters). See the Pipeline Components section below.
- Install BioPython (needed for tree manipulation).
- Install MOTHUR for 16S DNA alignments.
- Install FastTree for building 16S trees.
To configure RITA for rank-flexible classifications follow these steps:
- Follow the instructions for a rank-specific RITA installation given above.
- Update the MOTHUR and FastTree installation paths in globalsettings.cfg.
- Build a trusted BLAST database of 16S sequences. We recommend using the hand-curated sequences from RDP.
- Download and extract the unaligned Bacteria and Archaea sequences from RDP (link) into a directory called RDP:
> gunzip release10_28_unaligned.fa.gz
- Create the BLAST database:
> makeblastdb -in "release10_28_unaligned.fa" -dbtype nucl
- From the rita directory, BLAST the complete genomes against the 16S database:
> blastn -query ncbi_genomes.fna -db ../RDP/release10_28_unaligned.fa -out ncbi_genomes_16S.blast.txt -evalue 1e-10 -outfmt 6
- Use the get16s.py script to extract a single 16S sequence from each genome based on the best BLAST match:
> cd ./scripts > python get16s.py ../ncbi_genomes_16S.blast.txt ../../FCP/training/sequences ../../FCP/taxonomy.txt > mv sequences_of_16s.fasta ../
- The above script will produce the file sequences_of_16s.fasta which must be align. This can be done with MOTHUR using the following command:
mothur > set.dir(input=../rita) mothur > set.dir(output=../rita) mothur > align.seqs(candidate=sequences_of_16s.fasta, template=core_set_aligned.imputed.fasta, flip=t) mothur > quit()
- Place a copy of the 1349 character Lane Mask in your mothur directory.
- Note: core_set_aligned.imputed.fasta can be obtained from the MOTHUR here.
- Update the MOTHUR_16S_ALIGNMENT setting in globalsettings.cfg to point to the file sequences_of_16s.align which will be in the rita directory.
You are now ready to use rank-flexible RITA.
Rank-flexible example usage
To run rank-flexible RITA, you must first generate a proxy file for the 16S sequences contained in your sample:
python rita.py --buildproxy <sample_16s_fragments.fasta> --out proxy.txt
Then run rank-flexible RITA in the same way as rank-specific RITA, but specify the proxy file and the rank as FLEXIBLE
python rita.py --proxy proxy.txt --rank FLEXIBLE --pipeline NB_DCMEGABLAST,DCMEGABLAST_RATIO,NB_RATIO,NB_ML --query fragments.fasta --out results.txt
For more information on how rank-flexible RITA works, please see the publication.
Included pipeline components (labellers) (specify with --pipeline A,B,C,...)
NB_DCMEGABLAST - labels fragments that agree at rank X for NB and DCMEGABLAST NB_BLASTN - labels fragments that agree at rank X for NB and BLASTN NB_BLASTX - labels fragments that agree at rank X for NB and BLASTX NB_UBLASTX - labels fragments that agree at rank X for NB and UBLASTX DCMEGABLAST_RATIO - labels fragments that where the best DCMEGABLAST match evalue is at least Y times greater than the next best BLASTN_RATIO - labels fragments that where the best BLASTN match evalue is at least Y times greater than the next best BLASTX_RATIO - labels fragments that where the best BLASTX match evalue is at least Y times greater than the next best UBLASTX_RATIO - labels fragments that where the best BLASTX match evalue is at least Y times greater than the next best NB_RATIO - labels fragments that where the best NB match likelihood is at least Y times greater than the next best NB_ML - labels fragments that with the best NB match (if there are no ties) NULL_LABELLER - labels all remaining fragments with NONE (this should only ever be the last step in the pipeline).
rita.py accepts the following command-line parameters:
--help Provides a description of accepted command-line parameters. --pipeline Specify the components of the pipeline. --rank Taxonomic rank to classify at. --blastne BLASTN E-value threshold. --dblastne Discontiguous MegaBLASTN E-value threshold. --blastxe BLASTX E-value threshold. --ublastxe UBLASTX (usearch) E-value threshold. --blastnratio BLASTN E-value ratio. --dblastnratio Discontiguous MegaBLASTN E-value ratio. --blastxratio BLASTX E-value ratio. --ublastxratio UBLASTX (usearch) E-value ratio. --nb_ratio NB Likelihood ratio. --query FASTA file with query sequences. --out Output filename. --jobid Specify a job number. Default is a random 4 digit identifier. --buildproxy Build a proxy for rank-flexible classifications with the provided 16S sequences. --proxy Proxy for rank-flexible classifications created with --buildproxy.
Suggestions, comments, and bug reports can be sent to Rob Beiko (beiko [at] cs.dal.ca). If reporting a bug, please provide as much information as possible and a simplified version of the data set which causes the bug. This will allow us to quickly resolve the issue.
The development and deployment of RITA has been supported by several organizations:
- Genome Atlantic
- The Dalhousie Centre for Comparative Genomics and Evolutionary Bioinformatics, and the Tula Foundation
- The Natural Sciences and Engineering Research Council of Canada
- The Dalhousie Faculty of Computer Science