First Steps in the Logic-based Assessment of Post-composed Phenotypic Descriptions

Ernesto Jiménez-Ruiz, Bernardo Cuenca Grau, Rafael Berlanga Llavori, Dietrich Rebholz-Schuhmann

Introduction

Integrating diverse biological ontologies is a cornerstone of advanced medical reasoning. However, merging these complex systems often introduces unintended logical "bugs." This research presents a preliminary framework using Description Logic (DL) to automatically detect these unexpected consequences, ensuring that combined phenotypic descriptions remain medically accurate.

In bioinformatics, a phenotype is an observable characteristic. Traditionally, these are stored in "pre-composed" ontologies like the Human Phenotype Ontology (HPO), which function as flat dictionaries (e.g., "increased calcium concentration in blood"). The field is transitioning toward post-composed descriptions. Instead of a single flat term, a post-composed description links multiple specialized ontologies: a quality from PATO, a chemical from ChEBI, and an anatomical location from FMA.

Post-composed phenotype description for Hypercalcemia linking multiple ontologies

While this approach enables powerful automated reasoning, it creates a "clash of worlds." Because these ontologies were developed independently, merging them can lead to logical contradictions or "unintended consequences"—conclusions that are mathematically valid but medically nonsensical.

Methodology

The authors propose a logic-based assessment pipeline to identify these errors:

  1. Modularization: Reasoning over massive ontologies like the Foundational Model of Anatomy (FMA), which contains over 80,000 concepts, is computationally prohibitive. The authors utilized "locality-based modules" to extract relevant subsets, reducing the problem space to approximately 2,000 concepts.
  2. Reasoning: The HermiT reasoner was employed to classify the integrated ontology ($HPO_{ALL}$).
  3. Logical Difference: To identify changes post-integration, the team calculated the "logical difference" between the original HPO and the integrated version, highlighting new subsumption relationships.
  4. Approximation: Since exact logical difference calculation in complex DL is often undecidable, the researchers used approximations focusing on five specific types of simple logical axioms.

Datasets

The study integrated several prominent biomedical ontologies:

  • HPO (Human Phenotype Ontology): Primary source for phenotypic descriptions.
  • PATO (Phenotype and Trait Ontology): Qualities (e.g., "increased").
  • FMA (Foundational Model of Anatomy): Anatomical structures.
  • ChEBI (Chemical Entities of Biological Interest): Chemical compounds.
  • GO (Gene Ontology) & Cell Ontology: Biological processes and cell types.

Key Insights: The "Edema" Trap and Cross-Ontology Contamination

A significant finding of this study was the "Edema Trap." After integration, the system concluded that "Generalized edema" was equivalent to "Edema." This occurred due to the FMA's definition of body parts. While logically consistent within the integrated system, this is a medical error; generalized edema is a specific subtype, not a synonym.

Logical explanation for the unintended equivalence of Edema and Generalized Edema

Furthermore, the researchers identified "cross-ontology contamination." Axioms from one ontology (e.g., FMA's "part of" definition) could influence another (e.g., GO's biological processes), leading to erroneous conclusions such as classifying "Epidermis development" as an "Anatomical entity."

Despite these challenges, the study demonstrated that modularization makes logic-based assessment feasible. Reasoning time was reduced to just 45 seconds on a standard laptop, suggesting these methods can be integrated into real-time ontology development workflows.

Results and Future Directions

The integration triggered 759 new subsumption relationships within HPO concepts and discovered 80 new entailments linking concepts across different domain ontologies.

While this framework successfully identifies new logical consequences, it currently requires human experts to distinguish between valid discoveries and errors. Future work will focus on:

  • Developing systems to guide experts in "repairing" unintended consequences.
  • Expanding the tool to handle existing cross-references and mappings.
  • Adapting visualization tools like "ContentMap" to better assess complex integrations.

This research marks a critical first step in ensuring the logical integrity of the next generation of phenotypic descriptions.

Summary updated

Abstract

In this paper we present a preliminary logic-based evaluation of the integration of post-composed phenotypic descriptions with domain ontologies. The evaluation has been performed using a description logic reasoner together with scalable techniques: ontology modularization and approximations of the logical difference between ontologies.
BibTeX
@inproceedings{jimenezruiz2010first,
  author = {Ernesto Jiménez-Ruiz and Bernardo Cuenca Grau and Rafael Berlanga Llavori and Dietrich Rebholz-Schuhmann},
  title = {{First Steps in the Logic-based Assessment of Post-composed Phenotypic Descriptions}},
  booktitle = {SWAT4LS},
  year = {2010},
  url = {https://openreview.net/forum?id=Nhix4isdzR},
}
Figure