Skip to content
full taxonomer cython repository
C Perl Python
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.



Want to do get up and running with taxonomer quickly? Go to

metagenomics toolkit

Building the code:

This repository is written in c and python, with cython bringing the two together.

python dependencies -- we HIGHLY recommend using the anaconda distribution of python:

  1. Python 2.7+
  2. cython
  3. biopython

system dependency:

  1. openmp compatible C compiler

To build, cd into taxonomer/ -- directory where all the code lives, type the following commands:

  1. rm scripts/cython/*.so
  2. rm scripts/cython/*.c
  3. python build_ext --inplace

Building a nucleotide database:

Note that Taxonomer will ignore kmers with sequences in lower case

To build a database, you must create the kmer counts file with kanalyze v0.9.7, which is available here

The following is a command line example of kanalyze with the necessary paramaters (-k is the kmer size, -l and -d control the threading):

java -jar kanalyze.jar count -k 31 -d 3 -l 3 -o \<output file\> -m hex -rcanonical -f fasta \<input fasta\>

Then use to construct the database for classification (to see help, use python -h): <db_prefix> <kanalyze_output> <db_prefix.sti> <db_prefix.fa> -kl <kmer_length>

The fasta file needs to have sequence IDs as integer that match the numbers in the sti file. See below ("Creating taxonomic relationship files") for details about the sti and tri files. You may also use the script Utils/ on a fasta file with lineages in the description (see the script usage) to create them automatically from a fasta file.

Note that a random reference will be selected - if knowing the exact reference matters, consider asigning a taxid to each reference.

To parse the output of the classifier with a custom database, you may use the Util script

Building a protein database:

To build a protein database, you must follow these steps:

  1. Use to do necessary conversion before using kanalyze
  2. Run kanalyze to create kmer file, but with the following arguments (-k is the kmer size, no -rcanonical):
    java -jar kanalyze.jar count -k 30 -d 3 -l 3 -o \<output file\> -m hex -f fasta \<input fasta\>
  3. Use python, but specify --protein 1 to build protein database

Classifying reads

Use to classify reads. use -h argument to see command line options.
NOTE -- only use --protein 1 if you have already built a database for protein classification.

Binner database construction

We have empirically found binning using k-mers of 21bp in length to be effective for our research applications. This may or may not be the optimal size for your application, but the commands that follow demonstrate how to construct a binner database using 21bp k-mers.

  1. Create k-mer count files using kanalze of fasta files of reference nucleotide sequences:

java -jar kanalyze.jar count -k21 -d 3 -l 3 -rcanonical -o \<output file\> -m hex -f fasta \<input fasta\>

  1. Merge all k-mer count files using the script The help shows the following:

    positional arguments:

    kmer_counts_1 name of kmer count file in kanalyze hex output format kmer_counts_2 name of kmer count file to merge, must also be in kanalyze hex output format output merged output file

    optional arguments:

    -h, --help show this help message and exit -id_1 ID_1 integer value used to identify first database, must be a multiple of 2 -id_2 ID_2 integer value used to identify second databse, must be a multiple of 2

The first kmer count file, kmer_counts_1 will have kmer_counts_2 merged into it. Only use the -id_1 or -id_2 tags if the corresponding k-mer count file (kmer_counts_1 and/or kmer_counts_2) haven't already been merged using this script. The reason is that the flags -id_1, -id_2 will overwrite the id field in the file.

For example, suppose we have 3 kmer count files created by kanalyze to merge: k1.kc, k2.kc, k3.kc First, we determine the integer ids to assign to each kmer count file(must be a power of 2) -- n our example let k1.kc, k2.kc, k3.kc be 1, 2, 4 respectively. The commands to achieve this are (only two files can be merged at once):

python k1.kc k2.kc k1_k2.merged.kc -id_1 1 -id_2 2

python k1_k2.merged.kc k3.kc all_k.merged.kc -id_2 4

Now the merged file is called all_k.merged.kc

  1. Build our example database using :

python example_binner_db all_k.merged.kc fake.sti fake.fasta -kl 21 -binner

Where example_binner_db is the prefix for the output files that enable running the binner. fake.sti and fake.fasta are placeholders and are in the scripts/ folder. They can be used in any binner database build.

Creating taxonomic relationship files

Both building databases and classifying sequences requires knowlege about the taxonomic relationships between the organisms in the database. Taxonomer uses two files for this purpose, they have extensions .sti and .tri.

The .sti file is used during the build process. This file assigns every sequence id to a taxonomic id. It is highly recommended that sequence ids in fasta headers of the reference sequences be non negative integers and it is required that taxonomic ids be non negative integers. The .sti file is a modified fasta format with the sequence id in the header and the taxonomic id in the sequence line. For example, suppose we have the following fasta sequence:

>3 some interesting sequence

In the .sti file this sequence could be represented as follows (assuming its assigned taxid is 17):


The .tri file is used during classification. This file describes the parent-child relationship of the taxonomy. Like the .sti file, the .tri is a modified fasta format with the taxonomic child in the header and the taxonomic parent in the sequence line. It is required that the .tri be rooted at taxid '1' with 0 as its parent. Thus the following entry must be found in every .tri file:


The only retrictions imposed by taxonomer when defining a taxonomy is that every child has exactly 1 parent and the tree be rooted at taxid '1' with 0 as its parent. For this reason, it is possible to apply taxonomer effectively to a very wide range of mapping problems.

You can’t perform that action at this time.