Coding: Creating Mapping Testing Pipeline

[callahantiff]

GOALS: Create a pipeline that can be used to extend pediatric clinical-concept mappings to a new data source.

Workflow Update: @SteeleRobert has agreed to help with creating this pipeline.

Background:
We have created a large set of mappings from clinical diagnoses (n=29,128), medications (n=9,175 unique medications or 1,693 unique ingredients), and measurements (n=2,703 unique measurement results) to open biomedical ontologies.

TODO:

• Build a pipeline that performs multi-label classification.
  • Code should take in a set of OMOP codes and output mappings, with a confidence score to a specific set of ontologies:
    • Diagnosis codes ➞ Human Phenotype and Human Disease Ontologies
    • Medications ➞ NCBITaxon, Protein Ontology, ChEBI, and Vaccine Ontology
    • Measurement ➞ Human Phenotype
  • Can test code on UCHealth adult data set, MIMIC, and PIC

General Guidelines:

• Build scripts in an object-oriented framework
  • Sketch architecture before building code
  • General parsing class with subclasses by clinical type (i.e. conditions, measurements, medications)
• Test-driven development
• Needs to be written using keras and TensorFlow
• Inputs: a list of clinical codes, a list of ontologies

NEXT STEPS:

• Discuss this issue
• Agree on roles and authorship
• Discuss plan for moving forward, starting with discussions of architecture, prior to beginning coding

@SteeleRobert - are you good with this plan?

[SteeleRobert]

I was looking to this paper for information on multilabel classification. https://pdfs.semanticscholar.org/686d/f0720dc54c476dcb4d69584ef3c530d85f7f.pdf
I think where we can improve upon this is by trying to use the ontologies to improve clustering and ranking.

[callahantiff]

I was looking to this paper for information on multilabel classification. https://pdfs.semanticscholar.org/686d/f0720dc54c476dcb4d69584ef3c530d85f7f.pdf
I think where we can improve upon this is by trying to use the ontologies to improve clustering and ranking.

This looks really promising! I will make sure to read it prior to our meeting Friday so we can discuss it. Great find!