pypath: A Python module for molecular signaling prior knowledge processing


pypath supports both Python 2.7 and Python 3.6+. In the beginning, pypath has been developed only for Python 2.7. Then the code have been adjusted to Py3 and for a few years we develop and test pypath in Python 3. Therefore this is the better supported Python variant.



pypath consists of a number of submodules to build various databases. Most of these are provided as pandas data frames. The network database is built around igraph to work with molecular network representations e.g. protein, miRNA and drug compound interaction networks.


New webservice from 14 June 2018: the queries slightly changed, have been largely extended. See the examples below.

The webservice implements a very simple REST style API, you can make requests by the HTTP protocol (browser, wget, curl or whatever). After defining the query type and optionally a set of molecular entities (proteins) you can add further GET parameters encoded in the URL.

Query types

The webservice currently recognizes 7 types of queries: interactions, ptms, annotations, complexes, intercell, queries and info. The query types resources, network and about have not been implemented yet in the new webservice.

Interaction datasets

The instance of the pypath webserver running at the domain, serves not only the OmniPath data but also other datasets. Each of them has a short name what you can use in the queries (e.g. &datasets=omnipath,pathwayextra).

  • omnipath: the OmniPath data as defined in the paper, an arbitrary optimum between coverage and quality

  • pathwayextra: activity flow interactions without literature reference

  • kinaseextra: enzyme-substrate interactions without literature reference

  • ligrecextra: ligand-receptor interactions without literature reference

  • tfregulons: transcription factor (TF)-target interactions from DoRothEA

  • mirnatarget: miRNA-mRNA and TF-miRNA interactions

TF-target interactions from TF Regulons, a large collection additional enzyme-substrate interactions, and literature curated miRNA-mRNA interacions combined from 4 databases.

Mouse and rat

Except the miRNA interactions all interactions are available for human, mouse and rat. The rodent data has been translated from human using the NCBI Homologene database. Many human proteins do not have known homolog in rodents hence rodent datasets are smaller than their human counterparts. Note, if you work with mouse omics data you might do better to translate your dataset to human (for example using the pypath.homology module) and use human interaction data.


A request without any parameter provides the main webpage:

The info returns a HTML page with comprehensive information about the resources. The list here should be and will be updated as currently OmniPath includes much more databases:

Molecular interaction network

The interactions query accepts some parameters and returns interactions in tabular format. This example returns all interactions of EGFR (P00533), with sources and references listed.

By default only the OmniPath dataset used, to include any other dataset you have to set additional parameters. For example to query the transcriptional regulators of EGFR:

The TF Regulons database assigns confidence levels to the interactions. You might want to select only the highest confidence, A category:

Show the transcriptional targets of Smad2 homology translated to rat including the confidence levels from TF Regulons:

Query interactions from PhosphoNetworks which is part of the kinaseextra dataset:

Get the interactions from Signor, SPIKE and SignaLink3:

All interactions of MAP1LC3B:

By default partners queries the interaction where either the source or the arget is among the partners. If you set the source_target parameter to AND both the source and the target must be in the queried set:

As you see above you can use UniProt IDs and Gene Symbols in the queries and also mix them. Get the miRNA regulating NOTCH1:

Note: with the exception of mandatory fields and genesymbols, the columns appear exactly in the order you provided in your query.

Enzyme-substrate interactions

Another query type available is ptms which provides enzyme-substrate interactions. It is very similar to the interactions:

Is there any ubiquitination reaction?

And acetylation in mouse?

Rat interactions, both directly from rat and homology translated from human, from the PhosphoSite database:

Molecular complexes

The complexes query provides a comprehensive database of more than 22,000 protein complexes. For example, to query all complexes from CORUM and PDB containing MTOR (P42345):


The annotations query provides a large variety of data about proteins, complexes and in the future other kinds of molecules. For example an annotation can tell if a protein is a kinase, or if it is expressed in the hearth muscle. These data come from dozens of databases and each kind of annotation record contains different fields. Because of this here we have a record_id field which is unique within the records of each database. Each row contains one key value pair and you need to use the record_id to connect the related key-value pairs. You can easily do this with tidyr and dplyr in R or pandas in Python. An example to query the pathway annotations from SignaLink:

Or the tissue expression of BMP7 from Human Protein Atlas:

Roles in inter-cellular communication

Another query type is intercell providing information on the roles in inter-cellular signaling. E.g. if a protein is a ligand, a receptor, an extracellular matrix (ECM) component, etc. This query type is very similar to annotations but here the data does not come from original sources but combined from several databases by us. However we refer also to the original databases whenever the class_type is sub (subclass). E.g. the main class ligand is a combination of Ramilowski 2015, CellPhoneDB, HPMR and many other databases, hence besides the ligand category you will find sub-categories like ligand_ramilowski, ligand_cellphonedb and so on. An example how to get all intercell annotations for 4 selected proteins:

Or all the main classes for one protein:

Or a list of all ECM proteins:

Exploring possible parameters

Sometimes the names and values of the query parameters are not intuitive, even though in many cases the server accepts multiple alternatives. To see the possible parameters with all possible values you can use the queries query type. The server checks the paremeter names and values exactly against these rules and if any of them don’t match you will get an error message instead of reply. To see the parameters for the interactions query:

Can I use OmniPath in R?

You can download the data from the webservice and load into R. Thanks to our colleague Attila Gabor we have a dedicated package for this:

Alternatively here is a very simple example:



In almost any up-to-date Linux distribution the dependencies of pypath are built-in, or provided by the distributors. You only need to install a couple of things in your package manager (cairo, py(2)cairo, igraph, python(2)-igraph, graphviz, pygraphviz), and after install pypath by pip (see below). If any module still missing, you can install them the usual way by pip or your package manager.

igraph C library, cairo and pycairo

python(2)-igraph is a Python interface to use the igraph C library. The C library must be installed. The same goes for cairo, py(2)cairo and graphviz.

Directly from git

pip install git+

With pip

Download the package from /dist, and install with pip:

pip install pypath-x.y.z.tar.gz

Build source distribution

Clone the git repo, and run

python sdist

Mac OS X

On OS X installation is not straightforward primarily because cairo needs to be compiled from source. We provide 2 scripts here: the installs everything with HomeBrew, and installs from Anaconda distribution. With these scripts installation of igraph, cairo and graphviz goes smoothly most of the time, and options are available for omitting the 2 latter. To know more see the description in the script header. There is a third script which compiles everything from source and presumes only Python 2.7 and Xcode installed. We do not recommend this as it is time consuming and troubleshooting requires expertise.


  • no module named ... when you try to load a module in Python. Did theinstallation of the module run without error? Try to run again the specific part from the mac install shell script to see if any error comes up. Is the path where the module has been installed in your $PYTHONPATH? Try echo $PYTHONPATH to see the current paths. Add your local install directories if those are not there, e.g. export PYTHONPATH="/Users/me/local/python2.7/site-packages:$PYTHONPATH". If it works afterwards, don’t forget to append these export path statements to your ~/.bash_profile, so these will be set every time you launch a new shell.

  • pkgconfig not found. Check if the $PKG_CONFIG_PATH variable is set correctly, and pointing on a directory where pkgconfig really can be found.

  • Error while trying to install py(2)cairo by pip. py(2)cairo could not be installed by pip, but only by waf. Please set the $PKG_CONFIG_PATH before. See on how to install with waf.

  • Error at pygraphviz build: graphviz/cgraph.h file not found. This is because the directory of graphviz detected wrong by pkgconfig. See how to set include dirs and library dirs by --global-option parameters.

  • Can not install bioservices, because installation of jurko-suds fails. Ok, this fails because pip is not able to install the recent version of setuptools, because a very old version present in the system path. The development version of jurko-suds does not require setuptools, so you can install it directly from git as it is done in

  • In Anaconda, pypath can be imported, but the modules and classes are missing. Apparently Anaconda has some built-in stuff called pypath. This has nothing to do with this module. Please be aware that Anaconda installs a completely separated Python distribution, and does not detect modules in the main Python installation. You need to install all modules within Anaconda’s directory. does exactly this. If you still experience issues, please contact us.

Microsoft Windows

Not many people have used pypath on Microsoft computers so far. Please share your experiences and contact us if you encounter any issue. We appreciate your feedback, and it would be nice to have better support for other computer systems.

With Anaconda

The same workflow like you see in should work for Anaconda on Windows. The only problem you certainly will encounter is that not all the channels have packages for all platforms. If certain channel provides no package for Windows, or for your Python version, you just need to find an other one. For this, do a search:

anaconda search -t conda <package name>

For example, if you search for pycairo, you will find out that vgauther provides it for osx-64, but only for Python 3.4, while richlewis provides also for Python 3.5. And for win-64 platform, there is the channel of KristanAmstrong. Go along all the commands in, and modify the channel if necessary, until all packages install successfully.

With other Python distributions

Here the basic principles are the same as everywhere: first try to install all external dependencies, after pip install should work. On Windows certain packages can not be installed by compiled from source by pip, instead the easiest to install them precompiled. These are in our case fisher, lxml, numpy (mkl version), pycairo, igraph, pygraphviz, scipy and statsmodels. The precompiled packages are available here: We tested the setup with Python 3.4.3 and Python 2.7.11. The former should just work fine, while with the latter we have issues to be resolved.

Known issues

  • “No module fabric available.” – or pysftp missing: this is not important, only certain data download methods rely on these modules, but likely you won’t call those at all.

  • Progress indicator floods terminal: sorry about that, will be fixed soon.

  • Encoding related exceptions in Python2: these might occur at some points in the module, please send the traceback if you encounter one, and we will fix as soon as possible.

  • For Mac OS X (v >= 10.11 El Capitan) import of pypath fails with error: “libcurl link-time ssl backend (openssl) is different from compile-time ssl backend (none/other)”. To fix it, you may need to reinstall pycurl library using special flags. More information and steps can be found e.g. [here](

Special thanks to Jorge Ferreira for testing pypath on Windows!

Release History

Main improvements in the past releases:


  • First release of PyPath, for initial testing.


  • Lots of small improvements in almost every module

  • Networks can be read from local files, remote files, lists or provided by any function

  • Almost all redistributed data have been removed, every source downloaded from the original provider.


  • First version with partial Python 3 support.


  • pyreact module with BioPaxReader and PyReact classes added

  • Process description databases, BioPax and PathwayCommons SIF conversion rules are supported

  • Format definitions for 6 process description databases included.


  • Many classes have been added to the plot module

  • All figures and tables in the manuscript can be generated automatically

  • This is supported by a new module, analysis, which implements a generic workflow in its Workflow class.


  • chembl, unichem, mysql and mysql_connect modules made Python3 compatible


  • Orthology translation of network

  • Homologene UniProt dict to translate between different organisms UniProt-to-UniProt

  • Orthology translation of PTMs

  • Better processing of PhosphoSite regulatory sites


  • TF-target, miRNA-mRNA and TF-miRNA interactions from many databases


  • New web server based on pandas data frames

  • New module export for generating data frames of interactions or enzyme-substrate interactions

  • New module websrvtab for exporting data frames for the web server

  • TF-target interactions from DoRothEA


  • New dataio methods for Gene Ontology


  • Many new docstrings


  • New module complex: a comprehensive database of complexes

  • New module annot: database of protein annotations (function, location)

  • New module intercell: special methods for data integration focusing on intercellular communication

  • New module bel: BEL integration

  • Module go and all the connected dataio methods have been rewritten offering a workaround for data access despite GO’s terrible web services and providing much more versatile query methods

  • Removed MySQL support (e.g. loading mapping tables from MySQL)

  • Modules mapping, reflists, complex, ptm, annot, go became services: these modules build databases and provide query methods, sometimes they even automatically delete data to free memory

  • New interaction category in data_formats: ligand_receptor

  • Improved logging and control over verbosity

  • Better control over paremeters by the settings module

  • Many methods in dataio have been improved or fixed, docs and code style largely improved

  • Started to add tests especially for methods in dataio


  • The network database is not dependent any more on python-igraph hence it has been removed from the mandatory dependencies

  • New API for the network, interactions, evidences, molecular entities


  • New, more flexible network reader class

  • Full support for multi-species molecular interaction networks (e.g. pathogene-host)

  • Better support for not protein only molecular interaction networks (metabolites, drug compounds, RNA)


In the beginning the primary aim of pypath was to build networks from multiple sources using an igraph object as the fundament of the integrated data structure. From version 0.7 and 0.8 this design principle started to change. Today pypath builds a number of different databases each having pandas.DataFrame as a final format. Each of these integrates a specific kind of data from various databases (e.g. protein complexes, interactions, enzyme-PTM relationships, etc). pypath has many submodules with standalone functionality which can be used in other modules and scripts. For example the ID conversion module pypath.mapping.

Submodules perform various features, e.g. graph visualization, working with rug compound data, searching drug targets and compounds in ChEMBL.

ID conversion

The ID conversion module mapping can be used independently. It has the feature to translate secondary UniProt IDs to primaries, and Trembl IDs to SwissProt, using primary Gene Symbols to find the connections. This module automatically loads and stores the necessary conversion tables. Many tables are predefined, such as all the IDs in UniProt mapping service, while users are able to load any table from file or MySQL, using the classes provided in the module input_formats.


pypath includes data and predefined format descriptions for more than 25 high quality, literature curated databases. The inut formats are defined in the data_formats module. For some resources data downloaded on the fly, where it is not possible, data is redistributed with the module. Descriptions and comprehensive information about the resources is available in the descriptions module.

Structural features

One of the modules called intera provides many classes for representing structures and mechanisms behind protein interactions. These are Residue (optionally mutated), Motif, Ptm, Domain, DomainMotif, DomainDomain and Interface. All these classes have __eq__() methods to test equality between instances, and also __contains__() methods to look up easily if a residue is within a short motif or protein domain, or is the target residue of a PTM.


The module seq contains a simple class for quick lookup any residue or segment in UniProt protein sequences while being aware of isoforms.

Tissue expression

For 3 protein expression databases there are functions and modules for downloading and combining the expression data with the network. These are the Human Protein Atlas, the ProteomicsDB and GIANT. The giant and proteomicsdb modules can be used also as stand alone Python clients for these resources.

Functional annotations

GSEA and Gene Ontology are two approaches for annotating genes and gene products, and enrichment analysis technics aims to use these annotations to highlight the biological functions a given set of genes is related to. Here the enrich module gives abstract classes to calculate enrichment statistics, while the go and the gsea modules give access to GO and GSEA data, and make it easy to count enrichment statistics for sets of genes.

Drug compounds

UniChem submodule provides an interface to effectively query the UniChem service, use connectivity search with custom settings, and translate SMILEs to ChEMBL IDs with ChEMBL web service.

ChEMBL submodule queries directly your own ChEMBL MySQL instance, has the features to search targets and compounds from custom assay types and relationship types, to get activity values, binding domains, and action types. You need to download the ChEMBL MySQL dump, and load into your own server.


The module pypath.curl provides a very flexible download manager built on top of pycurl. The classes pypath.curl.Curl() and pypath.curl.FileOpener accept numerous arguments, try to deal in a smart way with local cache, authentication, redirects, uncompression, character encodings, FTP and HTTP transactions, and many other stuff. Cache can grow to several GBs, and takes place in ~/.pypath/cache by default. If you experience issues using pypath these are most often related to failed downloads which often result nonsense cache contents. To debug such issues you can see the cache file names and cache usage in the log, and you can use the context managers in pypath.curl to show, delete or bypass the cache for some particular method calls (pypath.curl.cache_print_on(), pypath.curl.cache_delete_on() and pypath.curl.cache_off(). You can always set up an alternative cache directory for the entire session using the pypath.settings module.

The pypath.session and pypath.log modules take care of setting up session level parameters and logging. Each session has a random 5 character identifier e.g. y5jzx. The default log file in this case is pypath_log/pypath-y5jzx.log. The log messages flushed in every 2 seconds by default. You can always change these things by the settings module. In this module you can get and set the values of various parameters using the pypath.settings.setup() and the pypath.settings.get() methods.

A simple webservice comes with this module: the server module based on twisted.web.server opens a custom port and serves plain text tables over HTTP with REST style querying.