kgml2sif

kgml2sif is a small command-line converter, written in Python 3 by Arnaud Poret (2019-2020) and released under the GPL-3.0 license, that transforms KEGG pathways encoded in KGML into the simple SIF tab-separated interaction format. KGML (KEGG Markup Language) is the XML representation of pathways distributed through the KEGG database and its REST API; SIF is the plain three-column "source, interaction, target" edge list historically associated with Cytoscape. kgml2sif exists to bridge these two worlds, turning KEGG's richly annotated XML graphics model into a lightweight edge list that network-analysis tools can ingest directly.

The tool reads one or more KGML files and emits a SIF file per input. It recognizes three node categories from KGML: genes (KEGG hsa: identifiers), compounds (cpd:, gl:, dr: prefixes), and groups, which represent molecular complexes. Complexes are flattened into synthetic nodes named by concatenating their components with an ampersand (for example cp1&cp2&cp3), and kgml2sif automatically emits "membership" edges linking each component to its complex. KEGG's typed relations are carried into the interaction column: PPrel (protein-protein), GErel (gene-expression regulation), PCrel (protein-compound), and ECrel (enzyme-enzyme relations that share a common compound). Relation subtypes such as activation, inhibition, or binding become the SIF interaction label; edges KEGG leaves unspecified are labelled "unknown."

Usage is deliberately minimal: kgml2sif [-g tsvFile] file.kgml .... The optional -g flag supplies a TSV lookup table to translate KEGG gene IDs into human-readable symbols (for example HRAS activation RAF1). During conversion the tool merges parallel multi-edges where compatible and prints warnings when it encounters conflicting relation types on the same node pair or KEGG identifiers it cannot match.

Its strengths are simplicity, transparency, and a clean mapping to a format that downstream graph software understands with no schema negotiation. The limitations follow from that minimalism: it targets human (hsa) pathways, covers only the four relation classes plus membership, discards KGML's layout, graphics, and much annotation, and produces an unlabelled flat edge list rather than a typed graph. For heavier, multi-format KEGG conversion the Java-based KEGGtranslator (KGML to SBML, BioPAX, GraphML, GML) or the R package KEGGgraph are more comprehensive alternatives; kgml2sif is best suited when a quick, scriptable KEGG-to-SIF step is all that is needed, for instance feeding pathway topology into NetworkX or Cytoscape.