Arachne: an OWL RL Reasoner Applied to Gene Ontology Causal Activity Models (and Beyond)

James P. Balhoff, Benjamin M. Good, Seth Carbon, Chris MungallISWC (P&D/Industry/BlueSky) 2018

Introduction

The Gene Ontology (GO) serves as the foundational framework for describing gene functions, biological processes, and cellular components. Traditionally, gene annotation involved simple "tagging." However, the GO Consortium is transitioning toward Causal Activity Models (GO-CAMs)—sophisticated semantic maps that detail the interactions between multiple gene products.

Building these complex models requires real-time logical validation. Curators need a reasoner capable of handling the massive GO "Tbox" (terminology), which contains nearly 1 million classes and 2 million axioms. Furthermore, the reasoner must support "Abox" (instance-level) reasoning, including inverse properties and property ranges, while operating fast enough to provide immediate feedback during the editing process. Existing tools like Apache Jena are insufficient at this scale, and others like ELK lack the necessary logic support.

Methodology: The Arachne Engine

Arachne is a high-performance, forward-chaining rule engine implemented in Scala, specifically designed to bridge the gap between massive ontologies and real-time curation.

Tbox-to-Rule Translation

The core innovation of Arachne is its approach to the Tbox. Rather than treating the ontology as passive data, Arachne uses an OWL API-based translator to convert Tbox axioms directly into executable logic rules. For instance, an axiom stating that every "nucleus" is an "organelle" is translated into a rule: (?x type nucleus) -> (?x type organelle).

The relationship between Tbox (ontology classes) and Abox (instance models) in GO-CAM, illustrating inferred types and relations.

Efficient Execution

Arachne utilizes the Rete/UL algorithm to efficiently match these rules against instance data. Upon startup, the engine pre-processes the million-plus rules into an immutable state. This design allows the system to reason over multiple GO-CAM models simultaneously in a multi-user environment without the overhead of complex locking mechanisms.

Evaluation and Results

The system was evaluated using the core components of the GO ecosystem, including the OBO Relations Ontology and ChEBI.

  • Scale: The GO Tbox comprises approximately 1 million classes and 2 million logical axioms.
  • Performance: On a standard 2017 MacBook Pro, Arachne pre-processes the full Tbox in roughly 60 seconds.
  • Real-time Reasoning: For a typical GO-CAM model consisting of hundreds of triples, Arachne materializes 1,000–2,000 inferred triples in just 1–2 seconds.
  • Comparative Advantage: In head-to-head testing, Arachne processed a GO-CAM dataset in ~1 second, whereas the Apache Jena reasoner required approximately 250 seconds.

Discussion and Future Work

Arachne effectively functions as a "logical spell checker," providing curators with immediate feedback by labeling inferred types and flagging errors as they occur. By "baking" the logic into rules before reasoning begins, it bypasses the performance bottlenecks typical of general-purpose reasoners.

Currently, Arachne is optimized for the OWL RL profile and does not yet support HasKey constructs or data properties (e.g., strings and numbers). Future work focuses on deeper integration with the Noctua modeling tool to provide proactive corrections and more intuitive explanations for its inferences.

Summary updated

BibTeX
@inproceedings{balhoff2018arachne,
  author = {James P. Balhoff and Benjamin M. Good and Seth Carbon and Chris Mungall},
  title = {{Arachne: an OWL RL Reasoner Applied to Gene Ontology Causal Activity Models (and Beyond)}},
  booktitle = {ISWC (P\&D/Industry/BlueSky)},
  year = {2018},
}
Figure