Skip to main content

tile phylogenetic space with subtrees

Project description

azulejo noun INFORMAL a glazed tile, usually blue, found on the inside of churches and palaces in Spain and Portugal.


azulejo azulejo combines homology and synteny information to tile phylogenetic space. The inputs to azulejo are FASTA files of nucleotide-space sequences of primary-transcript protein genes and their associated GFF files. Outputs are sets of proxy gene fragments chosen for their concordance in multiple sequence alignments, along with subtrees.


Python 3.7 or greater is required. azulejo is tested under Linux using Python 3.8 and 3.9 and under MacOS Big Sur using XCode command-line tools system Python (currently 3.8). Mac users should see the instructions on configuring their systems. Installation on BSD is not supported because many of the python dependencies lack BSD wheels.

We recommend you install azulejo into its own virtual environment due to the large number of python dependencies. The easiest way for most users to install and maintain up-to-date virtual environments is via the tool pipx. If your system does not have pipx installed, you can do so via the commands:

python3 -m pip install --user --upgrade pip
python3 -m pip install --user --upgrade pipx
python3 -m pipx ensurepath

Follow any instructions that the last command produces about starting a new shell if necessary.

If you choose to have azulejo compile and install its binary dependencies, you will need compilers, make, and cmake and standard headers for zlib and bz2. All linux systems configured for development will have these available. We test compilation under gcc version 10.2 on linux and clang 12.0.0 on MacOS. We use program-guided optimization for one of the binary dependencies, and we believe that gcc 10 does a much better job of optimization than gcc 9, so it may benefit you to upgrade your compiler if needed.

Installation for Users

Once the prerequisite has been met, you may then install azulejo in its own virtual environment by issuing the command:

pipx install azulejo

azulejo contains some long commands and many options. To enable command-line completion for azulejo commands, execute the following command if you are using bash as your shell:

eval "$(_AZULEJO_COMPLETE=source_bash azulejo)"

If you are using zsh, simply replace source_bash in the above command line with source_zsh.

You may next proceed with installing binary dependencies.

Environmental Variables

azulejo recognizes the following environmental variables:


This is a writable directory for installation of binary dependencies. Binaries will go into the bin directory. The default is the virtual environment directory.


This is the directory used for building binary dependencies. Default is the first memory device found for linux (e.g., /run/shm) or /tmp for MacOS. Set this if compilation fails because it runs out of memory.


This is the directory used for temporary merging of lists. The default is /tmp, but you may set it to a fast memory based device if you have enough memory.


These are the arguments to the make and make install commands when building dependencies. It’s good to set this to the number of processors on your system before installing required dependencies. This variable is only used during azulejo install.


This is the number of seconds between updates of the spinner. This defaults to 1, but it is advisable to set it higher for automated testing so as not to exceed logfile character limits.


If set to a log level such as info, the logger will be a simple print without using the more complex functions of loguru such as colors and logging to files. This is sometimes useful in automated testing.

In addition the optional dagchainer-tool subcommand recognizes the some environmental variables which can be shown via the command azulejo dagchainer-tool --help.

Installation of Binary Dependencies

azulejo requires MMseqs for homology clustering and MUSCLE for sequence alignment and initial tree-building. azulejo installs binaries into the virtualenv by default, so any systemwide installations of these packages will not get clobbered by the install. In particular, muscle is PGO-optimized, which gives nearly a factor of 2 higher performance than prebuilt binaries. For fastest installation, We recommand you set MAKEOPTS and install all required dependencies via the commands:

export MAKEOPTS=$(python -c 'import multiprocessing as mp; print(mp.cpu_count())')
azulejo install all

There are three optional dependencies that can be installed via azulejo install that are of interest only to a small subset of users who wish to compare against other homology clustering and synteny methods. usearch is a licensed homology clustering program that is free for individual, non-commercial use that can be downloaded and installed by the azulejo install usearch command after accepting the license terms. azulejo install dagchainer-tool gets you a somewhat crude Bash script that uses BLAST homology clustering followed by synteny calculation via DAGchainer. dagchainer-tool will need the dependency of perl with bioperl installed. dagchainer_tool increases the sequence ID length as part of its processing, so if any of your sequence IDS are longer than about 30 characters, they will violate BLAST’s hard limit of 50 characters in sequence ID fields. In that case you will need to install a patched version of BLAST using the command azulejo install blast-longids.

Installation For Developers

If you plan to develop azulejo, you’ll need to install the poetry dependency manager. If you haven’t previously installed poetry, execute the command:

curl -sSL | python

Next, get the master branch from GitHub

git clone

Change to the azulejo/ directory and install with poetry:

poetry install -v

Run azulejo with poetry:

poetry run azulejo


Installation puts a single script called azulejo in your path. The usage format is:


Master Input File

azulejo uses a configuration file in TOML format as the master input that associates files with phylogeny. The format of this file is the familiar headings in square brackets followed by configuration values:

rank = "genus"
name = "Glycine"

rank = "species"
name = "Glycine soja"

rank = "strain"
gff = "glyso.PI483463.gnm1.ann1.3Q3Q.gene_models_main.gff3.gz"
fasta = "glyso.PI483463.gnm1.ann1.3Q3Q.protein_primaryTranscript.faa.gz"
uri = ""
comments = """
Glycine soja accession PI 483463 has been identified as being unusually
salt-tolerant (Lee et al., 2009)."""
  • [headings]

    There can be only one top-level heading, and that will be the name of the resulting output set. This name will be the name of an output directory that will be created in the current working directory, so this heading (and all subheadings) must obey UNIX filesystem naming rules or an error will result. Each heading level (indicated by a “.”) will result in another taxonomic level and another directory level in the output directory. Depths do not need to be consistent.

  • rank

    Each level defined must have a rank defined, and that rank must match one of the taxonomic ranks defined by azulejo, which you can view and test using the check-taxonomic-rank command. There are 24 major taxonomic ranks, each of which may be modified by 16 different prefixes for a total of 174 taxonomic levels (some of which are synonoymous).

  • name

    Each level may (and usually should) have a name defined. This name is intended to be human-readable with no restrictions on the characters used, but it goes into plot legends in places, so it’s best to not make it too long. If the name is not specified, it will be taken from the level name enclosed in single quotes (e.g., ‘PI483463’ for the example above).

  • fasta

    If the level specifies a genome, it must have a fasta entry corresponding to the name of the protein FASTA file. In eukaryotes, the FASTA file should be a file of primary (generally longest) protein transcripts, if available, rather than all protein transcripts (i.e., not including splice variants). Sequences will be cleaned of dashes, stops, and other out-of-alphabet characters. Ambiguous residues at the beginnings and ends of sequences will be trimmed. Zero-length sequences will be discarded, which can result in a smaller number of sequences out. These files may be compressed, with extensions .gz or .bz2.

  • gff

    If the level specifies a genome, it must have a gff entry corresponding to a version 3 Genome Feature File (GFF3) containing CDS entries with ID values matching those IDs in the FASTA file. The same compression extensions as for fasta entries apply. If the SOURCE fields in those CDS entries (which contain the names of the DNA fragments such as scaffolds that the CDS came from) contain dot-separated components, those components that are identical across the entire file will be discarded by default. There is an opportunity later in the process to remap DNA source names to a common dictionary for comparison among chromosomes and plastids.

  • uri

    This optional field may contain a a uniform resource identifier such as https://sitename/dir/. azulejo uses smart-open for doing transparent on-the-fly decompression from a variety of file systems including HTTPS, HDFS, SSH, and SFTP (but not FTP). If this field is not supplied, local file access is assumed with paths relative to the current working directory. The URI will be prepended to fasta and gff paths, allowing for convenient downloading on-the-fly from sites such as LegumeInfo or GenBank. Downloads are not cached, so if you intend to run azulejo multiple times on the same input data, you will save time by downloading and uncompressing files to local storage.

  • preference

    This optional field may be used to override the genome preference heuristic that is the fall-thru preference after proxy-gene heuristics have been applied. This is an integer value, with lower integers getting the highest priority. Set this value to zero if you know in advance that one of the input genomes is considered the reference genome and, all things being equal, you would prefer to select proxy genes from this genome. You may also set these preference values later, after the default genome preference (genomes will be preferred in order of the most genes in a single DNA fragment) has already been applied, but before proxy gene selection.

  • other info

    A design goal for azulejo was to not lose metadata, even if it was not used by azulejo itself, while keeping metadata out of file names. As an aid in that goal, for each (sub)heading level/output directory, azulejo creates a JSON file named node_properties.json at each node in the output hierarchy that containing all information from this file as well as other information calculated at ingestion time by azulejo. You may specify any additional data you would like to pass along (e.g., for later use in a web page) and it will be translated from TOML to JSON and passed along, such as the multi-line comments field in the example. Examples of useful metadata that may be easier to enter at ingestion time than to garner later include taxon IDs of the level and its parent, common names, URLs of papers describing the genome, and geographic origin of the sample.

A copy of the input file will be saved in the output directory under the name input.toml. See the examples in the tests/testdata repository directory for examples of input data.

Global Options

The following options are global in scope and, if used must be placed before COMMAND:

-v, –verbose

Log debugging info to stderr.

-q, –quiet

Suppress logging to stderr.


Suppress logging to file.

-e, –warnings_as_errors

Treat warnings as fatal (for testing).


A listing of commands is available via azulejo --help. The currently implemented commands are, in the order they will normally be run:


Check for/install binary dependencies.


Marshal protein and genome sequence information.


Calculate homology clusters, MSAs, trees.


Calculate synteny anchors.


Calculate a set of proxy genes from synteny files.


Reads parquet file, writes tsv.

azulejo stores most intermediate results in the Parquet format with extension .parq. These binary files are compressed and typically can be read more than 30X faster than the tab-separated-value (TSV) files they can be interconverted with. In addition, Parquet files do not lose metadata such as binary representation sizes.

Each command has its COMMANDOPTIONS, which may be listed with:

azulejo COMMAND --help

Project Status

Latest Release

Python package

azulejo Definition


GitHub repository


Download stats

Download Rate

Download stats


License terms

Code Grade grade

Coverage test coverage

Travis Build

Travis CI


Issues reported

Code Style

Black is the uncompromising Python code formatter

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

azulejo-0.10.1.tar.gz (94.5 kB view hashes)

Uploaded source

Built Distribution

azulejo-0.10.1-py3-none-any.whl (98.0 kB view hashes)

Uploaded py3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page