HogProf

HogProf is an extensible and tunable approach to phylogenetic profiling using orthology data. It is powered by minhash-based data structures and computationally efficient.

Features

• Using orthoxoml files and a taxonomy calculated enhanced phylogenies of each family
• These are transformed into minhash signatures and a locally sensitive hashing forest object for search and comparison of profiles
• Taxonomic levels and evolutionary event types (presence, loss, duplication) can have custom weight in profile construction
• Optimization of weights using machine learning

If you run into any problems, feel free to contact me at dmoi@unil.ch

Quickstart

Install from PyPI (recommended)

pip install hogprof

Install from sources

git clone https://github.com/DessimozLab/HogProf.git
cd HogProf
pip install .

Usage: main mode

Example: using the OMA database

Let's get a current version of the OMA hdf5 file and GAF. This will aloww us to use the HOGs and study the functional enrichment of our search results. Careful, this download is heavy (hundreds of GB):

mkdir YourOmaDirectory
cd YourOmaDirectory
wget https://omabrowser.org/All/OmaServer.h5
wget https://omabrowser.org/All/oma-go.txt.gz

The latest release (May.2026) is more than 300GB in size. Alternatively, you can use earlier releases, for example:

wget https://omabrowser.org/All.Jul2024/OmaServer.h5

which is 180GB in size. For other releases, check https://omabrowser.org/oma/archives/, select release and search for "OMA Browser database (as hdf5)".

We also need to make a location to store our pyprofiler databases

cd ..
mkdir YourHogProfDirectory

Let's now compile a database containing all HOGs and our desired taxonomic levels using default settings. Launch the lshbuilder as shown below. dbtypes available on the command line are: all, plants, archaea, bacteria, eukarya, protists, fungi, metazoa, and vertebrates. These will use the NCBI taxonomy as a tree to annotate events in different gene family's histories.

If you are using an OMA release before 2022 you will need to use the NCBI tree. This is the default tree used by HogProf.

lshbuilder --outpath YourHogProfDirectory --dbtype all --OMA YourOmaDirectory/OmaServer.h5 --nthreads numberOfCPUcores         

If you are using OMA releases after 2022 we will also be downloading the OMA taxonomic tree since it is more accurate than the NCBI tree. This will be used to annotate the events in the gene family histories.

wget https://omabrowser.org/All/speciestree.nwk

We are ready to use the OMA tree to build our database.

lshbuilder --outpath YourHogProfDirectory --dbtype all --OMA YourOmaDirectory/OmaServer.h5 --nthreads numberOfCPUcores --mastertree YourOmaDirectory/speciestree.nwk --reformat_names True

This should build a taxonomic tree for the genomes contained in the release and then calculate enhanced phylogenies for all HOGs in OMA.

Once the database is completed it can be interogated using a profiler object. Construction and usage of this object should be done using a python script or notebook. This shown in the example notebook searchenrich.ipynb found in the examples. Please feel free to modify it to suit the needs of your own research.

Usage: levels mode

The installation process is the same (github only for now).

A fastOMA result folder is expected as input.

Step 1: split the orthoxml into HOG-specific orthoxmls, using this script (requires a fastOMA installation):

python path/to/HogProf/src/HogProf/splitter.py \
    /path/to/output_directory \
    --fastoma-path /path/to/FastOMA

Step 2: build the HogProf database in levels mode by running lshbuilder.py with the --slicesubhogs option enabled. Point --OrthoGlob to the HOG-specific OrthoXML files generated in Step 1 and provide the species tree used by FastOMA.

python path/to/HogProf/src/HogProf/lshbuilder.py \
    --OrthoGlob "/path/to/fastoma_results/splits/*.orthoxml" \
    --outpath "/path/to/output_database/" \
    --mastertree "/path/to/fastoma_results/species_tree_checked.nwk" \
    --reformat_names True \
    --slicesubhogs True \
    --eventslim 1 \
    --nthreads 20

Required parameters:

• --OrthoGlob: glob matching the HOG-specific OrthoXML files created in Step 1.
• --outpath: output directory for the HogProf database.
• --mastertree: species tree corresponding to the FastOMA run.
• --slicesubhogs True: enables levels mode by generating profiles for subHOGs.

Common optional parameters:

• --eventslim: minimum number of evolutionary events required to retain a subHOG (e.g. 1).
• --specieslim: minimum number of species required in a subHOG (default: 10).
• --nthreads: number of CPU threads to use.
• --reformat_names True: recommended if the species tree labels need to be converted to numeric identifiers.

Step 3: generate Jaccard similarity distributions and compute threshold tables from the HogProf database.

This step analyzes the output of lshbuilder and produces the bin-wise Jaccard threshold table used for downstream filtering and scoring.

python path/to/HogProf/src/HogProf/generate_jaccard_distr.py \
    --hogprof_folder "/path/to/2a_hogprof_output/" \
    --output "/path/to/output_distributions/" \
    --profiler-path "/path/to/profiler.py"

Required parameters:

• --hogprof_folder: folder containing the HogProf LSHbuilder output (Step 2).
• --output: directory where distribution plots and threshold tables will be written.

Optional parameters:

• --profiler-path: path to profiler.py.
  By default, this points to the profiler.py located in the same directory as the script.

Output

This step produces a key file used for thresholding:

• bins_jaccard_thresholds.csv
  → table of Jaccard similarity thresholds computed per bin, used for downstream scoring and filtering.

The file is saved in the directory specified by --output.

Step 4: extract level-specific hits using the HogProf profiler.

This step runs extract_levels_hits.py to query the HogProf database built in Step 2 and generate a ranked list of hits per subHOG using the threshold table generated in Step 3.

python path/to/HogProf/src/HogProf/extract_levels_hits.py \
    --input "/path/to/hogprof_output/" \
    --outputfile "/path/to/output/extracted_hits.csv" \
    --queries_file "/path/to/queries.tsv" \
    --thresholds "/path/to/bins_jaccard_thresholds.csv" \
    --k 1000 \
    --allvsall True \
    --profilerpath "profiler.py"

Required parameters:

• --input: HogProf LSHbuilder output directory (Step 2).
  Must contain files such as: newlshforest.pkl, hashes.h5, reformatted_tree.nwk, fam2orthoxml.csv, profilersavingpath.csv.
• --outputfile: path to the CSV file where extracted hits will be saved.
• --queries_file: table of query subHOGs (e.g. venom HOGs).
  Must contain a hashid column and a subHOG identifier column.
• --thresholds: CSV file generated in Step 3 (bins_jaccard_thresholds.csv).

Optional parameters:

• --k: number of top hits to retain per query (default: 1000).
• --allvsall: also performs all-vs-all comparisons (default: True, may be computationally expensive).
• --profilerpath: path to profiler.py from HogProf (default assumes local availability).

Output

This step produces:

• extracted_hits.csv
  → ranked hit table containing similarity matches between query subHOGs and the HogProf database, filtered using bin-specific Jaccard thresholds.

Step 5: visualize coevolution networks in Cytoscape.

This final step uses the Jupyter notebook cytoscape_import.ipynb to generate and export coevolution networks from the extracted HogProf hits.

The notebook loads the output from Step 4 and prepares network files that can be visualized in Cytoscape.

In addition to the extracted hits, the notebook requires a nodes metadata table corresponding to the query subHOGs used in Step 4.

Once the database is completed it can be interogated using a profiler object. Construction and usage of this object should be done using a python script or notebook. This shown in the example notebook searchenrich.ipynb found in the examples. Please feel free to modify it to suit the needs of your own research.