PAX2GraphML

PAX2GraphML is an open-source Python library for converting biological pathway data expressed in BioPAX into graph representations, so that large-scale regulation, signaling and metabolic networks can be manipulated with standard graph tooling. It was developed by François Moreews, Hugo Simon, Anne Siegel, Florence Gondret and Emmanuelle Becker (Univ Rennes / Inria / CNRS / IRISA and INRAE Pegase, France) and published in Bioinformatics in 2021. The project is distributed on PyPI under the MIT license, with a Docker image and a hosted resource site at pax2graphml.genouest.org.

The library's central abstraction is the "regulated reaction," implemented as the SPAIM model (Substrate, Product, Activator, Inhibitor, Modulator). Rather than preserving the many distinct BioPAX class types, it collapses biochemical reactions and regulatory interactions into a homogeneous node-and-edge form where each reaction node connects to its inputs (substrates, modulators such as activators and inhibitors) and outputs (products), with BioPAX aliases and metadata carried onto GraphML node and edge properties. This lets regulatory, signaling and metabolic levels be represented uniformly in a single graph. It reads the OWL-based BioPAX and emits GraphML, the same interchange format used across the wider graph ecosystem.

Internally, PAX2GraphML relies on the Java library Paxtools to parse BioPAX and extract reaction patterns (requiring a JVM), and on the C++-backed graph-tool library for efficient in-memory graph construction and analysis of the resulting GraphML. It can build graphs from a single BioPAX source or combine and filter several sources, supporting sub-graph extraction, database merging, classical metrics (degree, betweenness, closeness, connected components) and generation of signed influence graphs. Because output is plain GraphML, downstream work can continue in NetworkX, igraph or R libraries, or be visualized in Cytoscape or Gephi.

The authors demonstrate scale by processing all 16 Pathway Commons datasources into a single 363 MB GraphML graph on a 48 GB machine, a run that took about seven days, so the pipeline is batch-oriented rather than interactive at full scale, and pre-built database exports are offered to avoid repeating it. Practical limitations include the heavyweight dependency stack (Java plus graph-tool, which is nontrivial to install), a narrow BioPAX-to-GraphML scope with no reverse or broader format conversion, and a small, research-driven maintenance footprint. It occupies a niche similar to Paxtools and paxtoolsr but is aimed specifically at turning curated pathway databases into analyzable regulation graphs.