Topology and redescriptions detect multiple alternative biological pathways from clinical phenotypes

Area of Science:

• Computational biology
• Bioinformatics
• Data science in medicine

Background:

• Biological pathways are critical for understanding disease mechanisms and drug discovery.
• Extracting pathway information from unstructured clinical notes is challenging due to data complexity and potential physician bias.

Purpose of the Study:

• To develop an efficient computational pipeline for identifying disease-associated biological pathways.
• To address the challenges of unstructured electronic health records and physician bias in phenotype extraction.

Main Methods:

• Utilized topological data analysis and redescriptions for knowledge extraction.
• Developed a four-phase pipeline: concept extraction, feature space construction, topological analysis, and bias investigation.
• Applied the pipeline to a dataset of COVID-19 clinical notes.

Main Results:

• The pipeline successfully extracted salient concepts and constructed patient feature spaces.
• Topological analysis distilled knowledge and visualized extracted features.
• Identified meaningful biological pathways associated with diseases from clinical notes.

Conclusions:

• The proposed pipeline offers an efficient approach to uncover disease-related pathways from unstructured clinical data.
• This method can aid practitioners in understanding disease mechanisms and inform drug discovery efforts.
• The study demonstrates the utility of topological data analysis in biomedical research.