Abarenkov, K., Somervuo, P., Nilsson, H., Kirk, P., Huotari, T., Abrego, N. and Ovaskainen, O. 2018. PROTAX-fungi: a web-based tool for probabilistic taxonomic placement of fungal ITS sequences. New Phytologist, doi: 10.1111/nph.15301.
The results of PROTAX-fungi include interactive KRONA-charts showing the classifications and classification reliabilities of the environmental sequences. Here such charts are exemplified for case studies on root-associated fungi (greenland90.html) and for wood-associated fungi (sawdust90.html) considered by Abarenkov et al. (2018). They can be accessed by downloading and unzipping protaxexamples.zip.
After running PROTAX for the set of input sequences, all results are in a single output directory. Predicted class labels are in text files and output directory contains an interactive HTML file where the classifications are shown in a taxonomy pie chart.
The main output is the set of 6 text files which contain the taxonomic classifications for 6 levels from phylumn to species:
| Filename | Description |
|---|---|
| query2.nameprob | Taxonomic classifications in phylum level (seqID taxon probability) |
| query3.nameprob | Taxonomic classifications in class level (seqID taxon probability) |
| query4.nameprob | Taxonomic classifications in order level (seqID taxon probability) |
| query5.nameprob | Taxonomic classifications in family level (seqID taxon probability) |
| query6.nameprob | Taxonomic classifications in genus level (seqID taxon probability) |
| query7.nameprob | Taxonomic classifications in species level (seqID taxon probability) |
Each line of the file has the following format:
seqID taxon1 probability1 taxon2 probability2 ... taxonN probabilityN
All taxa whose probabilities exceeds 0.01 are listed for each input sequence. Taxon names include the path from kingdom level (Fungi) to the level (phylum, class, order, family, genus, species) of the file. The order of input sequences is the same in all 6 files.
Krona HTML file 'krona.html' can be interactively visualized with web browser. Piechart shows all those taxonomic units for which there are classifications. Classification confidence is shown when user clicks the box "Color by Confidence". The six confidence categories for taxonomic unit with reference sequences (TunitA) and taxonomic unit for which there are no reference sequences (TunitB) are:
- TunitA at least half of the input sequences classified to this taxon have classification probability > th
- TunitA less than half of the input sequences classified to this taxon have classification probability > th
- TunitA none of the input sequences classified to this taxon have classification probability > th
- TunitB at least half of the input sequences classified to this taxon have classification probability > th
- TunitB less than half of the input sequences classified to this taxon have classification probability > th
- TunitB none of the input sequences classified to this taxon have classification probability > th
where th is the threshold for probability given by user (e.g. 0.9). Note that this threshold is only used for coloring the pie chart, it does not affect how PROTAX reports the classification probabilities in the query.nameprob files. TunitB contains both the case where there are no reference sequences to known taxonomic unit and the case where the taxonomic unit is outside of the known taxonomy (unknown branch).
By default, each input sequence represents a single sequence but it can also represent a cluster of sequences. Cluster size information can be included in the FASTA ID of the input sequence. This does not affect the classification of the sequence, but it affects the sizes of the taxa in the pie chart. For details, see the end of this README file.
| Filename | Description |
|---|---|
| krona.xml | XML input file used to create Krona pie chart |
| query1.logprob | Initial taxonomic placement of each sequence in kingdom level (fixed to be Fungi) |
| query2.logprob | Taxonomic classifications with phylum node indices as output |
| query3.logprob | Taxonomic classifications with class node indices as output |
| query4.logprob | Taxonomic classifications with order node indices as output |
| query5.logprob | Taxonomic classifications with family node indices as output |
| query6.logprob | Taxonomic classifications with genus node indices as output |
| query7.logprob | Taxonomic classifications with species node indices as output |
| query.fa | FASTA input sequence file |
| query.ids | Sequence identifiers of input sequences |
| query.lensize | Cluster sizes and sequence lengths of input sequences |
| query.m8 | usearch_global output (queryID referenceID sequenceSimilarity) |
| query.sasintax | SINTAX output formatted to PROTAX |
| query.sintax | original SINTAX output (queryID taxon,probability list) |
Sections below contain information for administrator of PROTAX server and they are not necessarily relevant for user
Extract files from file 'protaxfungi.tgz' (tar xvfz protaxfungi.tgz). This will create directory 'protaxfungi' which contains PROTAX-Fungi version 1 (without sequence data and taxonomy files). Subdirectories under 'protaxfungi' are:
- model: taxonomy, reference sequences, and model parameters
- protaxscripts: Perl scripts needed for running PROTAX
- thirdparty: external tools to calculate sequence similarity predictors and make Krona pie charts
Things to set before running PROTAX:
Directory 'model' contains files with the name 'paramsR.levelL', where R is 'its1', 'its2', or 'itsfull'. For each sequence region there are parameters for all taxonomy levels, i.e. L=2..7 (level 1 represents kingdom but it is excluded since the classification starts directly from Fungi node in kingdom level). Levels are 2:phylum, 3:class, 4:order, 5:family, 6:genus, and 7:species. Model parameters are trained for UNITE sequence data and taxonomy using USEARCH version usearch10.0.240_i86linux32.
Put sequence data and taxonomy tree in the directory 'model'. The needed files are:
- reference sequences in FASTA format for each sequence region: its1.fa, its2.fa, itsfull.fa. When models have been trained, ITS1 and ITS2 regions have been extracted using ITSx software and only those sequences have been used which represent full ITS region, i.e. both ITS1 and ITS2 have been extracted from the original sequence.
- ref.tax[2..7] files contain sequence ID and taxon name pairs for each taxonomy level (one line per sequence ID)
- rseqs[2..7] files contain the list of reference sequence IDs for each taxonomy node (one line per taxon)
- seqid2tax file contains full taxonomy path for each sequence ID
- taxonomy file contains taxonomy tree with all levels
- tax[L=2..7] files contain taxonomy tree for each level separately
- sintax predictor related files:
- seqid2tax.ascii7: seqid2tax file where special characters have been replaced
- taxonomy.ascii7: taxonomy file where special characters have been replaced
- sintax[its1,its2,itsfull]train.fa: reference data suitably formatted for sintax
2) Sequence similarity related predictors are based on 'usearch_global' and 'sintax' from USEARCH package. Download USEARCH from https://www.drive5.com/usearch/download.html
4) Make reference sequences available to USEARCH by giving the following commands in directory 'model':
$ cd model
$ ../thirdparty/usearch10.0.240_i86linux32 -makeudb_usearch its1.fa -output its1.udb
$ ../thirdparty/usearch10.0.240_i86linux32 -makeudb_usearch its2.fa -output its2.udb
$ ../thirdparty/usearch10.0.240_i86linux32 -makeudb_usearch itsfull.fa -output itsfull.udb
$ ../thirdparty/usearch10.0.240_i86linux32 -makeudb_sintax sintaxits1train.fa -output sintaxits1.udb
$ ../thirdparty/usearch10.0.240_i86linux32 -makeudb_sintax sintaxits2train.fa -output sintaxits2.udb
$ ../thirdparty/usearch10.0.240_i86linux32 -makeudb_sintax sintaxitsfulltrain.fa -output sintaxitsfull.udb
5) Download Krona from https://github.com/marbl/Krona/wiki
$ cd thirdparty
$ tar xvf KronaTools-2.6.1.tar
$ cd KronaTools-2.6.1
$ perl install.pl --prefix ../krona
and make Krona available in PATH (this is for Bash script 'runprotax'):
$ export PATH=$PATH:pwd/krona/bin
7) Set variable PROTAXDIR correctly in Bash script 'runprotax'. It should be the path to the present directory where the subdirectories 'model', 'protaxscripts' and 'thirdparty' are.
Bash script 'runprotax' is used for running PROTAX using either ITS1, ITS2, or ITSfull model. ITSfull includes the entire ITS1_5.8S_ITS2 region.
Two shell variables ITS and ODIR need to be set before running the script. Variable ITS defines the sequence model, it needs to be its1, its2, or itsfull.
Sequences to be classified should be in FASTA format in a file named 'query.fa'. Before running PROTAX, user should set variable ODIR to be the parent directory of this file. PROTAX will write all its output in this directory. E.g. if the original FASTA file is in ~/allseqs/sample1.fa and the sequences represent ITS2 region, the following commands will create a new directory for this sample and run PROTAX for it:
$ ODIR=~/myprotaxout/sample1
$ ITS=its2
$ mkdir $ODIR
$ cp ~/allseqs/sample1.fa $ODIR/query.fa
$ source runprotax
After runprotax is finished, PROTAX classifications and Krona piechart are in $ODIR. Outputs from each taxonomy level are in separate files:
- phylum classifications in $ODIR/query2.nameprob
- class classifications in $ODIR/query3.nameprob
- order classifications in $ODIR/query4.nameprob
- family classifications in $ODIR/query5.nameprob
- genus classifications in $ODIR/query6.nameprob
- species classifications in $ODIR/query7.nameprob
Note: when creating Krona piechart, if input FASTA file has ';' in ID lines, the 2nd column is interpreted as cluster size which sequence represents. Size information may also include prefix 'size='. Here are four examples of FASTA IDs:
- >seqidA => cluster size = 1 (since no other information given)
- >seqidB;size=150 => cluster size = 150
- >seqidC;40 => cluster size = 40
- >seqidD;fungi => cluster size = 1 (since 'fungi' is not a number)