Biomedical Knowledge Graphs Construction From Conditional Statements | JoVE Visualize

Area of Science:

Biomedical Informatics
Knowledge Representation
Computational Biology

Background:

Existing biomedical knowledge graphs (BioKGs) primarily focus on factual knowledge, neglecting the crucial contextual conditions under which these facts are valid.
This omission leads to a loss of essential context for knowledge exploration and inference in biomedical statements.
There is a need for a more comprehensive BioKG structure that incorporates both facts and their associated conditions.

Purpose of the Study:

To propose a novel three-layered, information-lossless representation for biomedical knowledge graphs (BioKGs) that integrates both factual information and its conditions.
To develop a robust methodology for constructing these enhanced BioKGs by framing the problem as a sequence labeling task.
To evaluate the performance of the proposed model against existing methods for BioKG construction.

Main Methods:

A three-layered BioKG structure was designed, comprising concept/attribute nodes, fact/condition tuple nodes, and statement nodes.
The BioKG construction problem was transformed into a sequence labeling task using a novel tag schema.
A Multi-Input Multi-Output (MIMO) sequence labeling model was developed to learn from multiple input signals and generate multiple output sequences for tuple extraction.

Main Results:

The proposed MIMO model demonstrated superior performance compared to existing methods in constructing BioKGs on a newly created dataset.
The constructed BioKGs, incorporating conditions, provided a richer and more nuanced understanding of biomedical statements.
Case studies confirmed the practical utility of the enhanced BioKG representation for biomedical knowledge exploration.

Conclusions:

The proposed three-layered BioKG representation effectively captures both facts and their conditions, addressing limitations of existing flat BioKGs.
The MIMO sequence labeling model offers an efficient and effective approach for constructing these context-aware BioKGs.
This work advances the field of biomedical informatics by enabling more comprehensive knowledge representation and facilitating deeper insights from biomedical data.