Analyzing Inconsistency Toward Enhancing Integration of Biological Molecular Databases

Yi-Ping Phoebe Chen, Qingfeng ChenAPBC 2006 · OpenReview

Abstract

The rapid expansion of biological databases has created significant challenges for data integration due to structural and terminological heterogeneity. This paper proposes a formal mathematical framework that utilizes ontologies to quantify semantic inconsistency between disparate databases. By introducing a "compatibility" metric, the framework allows researchers to evaluate and filter data sources prior to integration, thereby improving the reliability of downstream bioinformatics and data mining processes.

Introduction

Modern life sciences rely heavily on massive repositories such as GenBank or NCBI to investigate genomes, protein structures, and evolutionary biology. Answering complex biological questions often necessitates querying multiple databases simultaneously. However, these databases are frequently developed by independent teams for diverse purposes, leading to significant structural and terminological heterogeneity. For instance, one database may utilize common English names for species, while another employs strict systematic binary formats.

While ontologies (standardized, hierarchical vocabularies like the Gene Ontology) are increasingly used to bridge these gaps, the underlying inconsistency between databases is often overlooked. Blindly merging highly discrepant databases introduces noise and unreliability into datasets, potentially compromising downstream machine learning or data mining efforts. This research addresses this problem by establishing a metric to quantify database inconsistency before integration.

Methodology

The proposed framework utilizes ontologies as an auditing tool rather than just a linking mechanism. The core of the methodology involves several semantic operators:

  1. Mapping: Linking a database column to a specific ontology concept.
  2. Cross-reference: Connecting two different databases via a shared concept.
  3. Translation: Managing synonyms and different naming conventions.
  4. Taxonomy: Navigating hierarchical relationships such as "is-a" or "part-of."
Figure 1: Mapping database attributes to an ontology hierarchy.

When a user queries an attribute, the system maps it to an ontology concept and compares a reference database against target databases. The algorithm categorizes attributes into a Consistentset (semantic matches) and a Conflictset (conflicting or null values). The compatibility is then calculated using the formula:

Compatibility = |Consistentset| / (|Consistentset| + |Conflictset|)

If the resulting score falls below a user-defined threshold (e.g., 0.5), the databases are considered too inconsistent for safe integration regarding that specific query.

Experimental Setup

To demonstrate the framework, the authors constructed five hypothetical biological databases (DB1 through DB5) modeled after real-world repositories. These databases contain fields such as DNA Sequence, Description, Identifier, Organism, and Enzyme.

Figure 3: Structure of the five hypothetical biological databases used for testing.

Results and Discussion

The framework was evaluated through qualitative proof-of-concept scenarios:

  • Scenario 1 (Enzyme Query): A query for the attribute enzyme: mouse across the five databases identified one consistent match and three conflicting/null matches, resulting in a compatibility score of 0.25. This successfully flagged the databases as inconsistent for this specific query.
  • Scenario 2 (Taxonomic Sensitivity): A query for animal: mouse highlighted the system's sensitivity to the reference attribute. Using a general organism (org) attribute yielded a compatibility score of 0.5, whereas a highly specific systematic species name (spec) attribute dropped the score to 0.25.

These results underscore that consistency is relative to the specific query and chosen reference. Ontologies serve as a mathematical "measuring stick" to calculate semantic distance and conflict.

Conclusion and Future Work

Measuring inconsistency is a vital prerequisite for database integration to avoid "garbage in, garbage out" scenarios. While achieving perfect compatibility (1.0) is often impossible due to inherent discrepancies in biological nomenclature, this framework provides a foundational step for automated data pipelines. Future work should focus on evaluating the framework's scalability on massive, live biological databases to enable dynamic filtering of low-quality data sources in real-time knowledge discovery.

Summary updated

BibTeX
@inproceedings{chen2006analyzing,
  author = {Yi-Ping Phoebe Chen and Qingfeng Chen},
  title = {{Analyzing Inconsistency Toward Enhancing Integration of Biological Molecular Databases}},
  booktitle = {APBC},
  year = {2006},
  url = {https://openreview.net/forum?id=rKZmkJjPUp},
}
Figure