Programming Question

A core challenge in digital health/biomedical informatics is to take the complex, continuous world of healthcare and somehow represent it in a way that computers can assist in better decision making. In Assignment 1 you tackled one aspect of this task: taking healthcare experiences and translating these into computable data representations. In this assignment we tackle the complementary task of representing biomedical knowledge in a computable format.

As Davis, Shrobe, and Szolovits pointed out “knowledge representation is one of the central and…concepts in AI” [@Davis_Shrobe_Szolovits_1993]. And as Homer Warner realized way back in the 1960s, without adequate knowledge representation healthcare workers will simply be overwhelmed by the larger and larger amounts of digital data we collect on patients (see the materials in the first module of this subject).

In this assignment you will create a knowledge representation for a particular disease or health condition. Specifically, you will make a knowledge graph. In order to leverage your work in Assignment 1, you should create the knowledge graph for the same domain you used for Assignment 1. (However, if you are absolutely sick of that topic, you can move on to a different problem.)

A knowledge graph is simply a

directed graphLinks to an external site.

where meaning is given to each node (vertex) and edge (arrow) in the graph.

Your knowledge graph should provide a reasonable representation of your chosen problem. Each concept (node/vertex) in the graph should, if possible, be defined by reference to a concept in an appropriate ontology (e.g.

SNOMED-CTLinks to an external site.

,

ICD-10Links to an external site.

,

FMALinks to an external site.

). All definitions should be referenced by a URL (e.g.

https://radlex.org/RID/RID4553Links to an external site.

). If you cannot find your concept in an appropriate ontology, you can create your own definition of the concept. Your definition needs to be available online (e.g. a Gist on GitHub, a Google Doc, etc.); your self-created definitions must follow the principles of concept-based representation: a clearly defined concept, a human label, and a non-semantic unique identifier). Your concept definitions do not need to be from the same ontology.

Similarly, the meaning assigned to each edge should be defined by reference to an external standard. For example,

schema.orgLinks to an external site.

defines the following relationships that may be relevant to your knowledge graph:

causeOfLinks to an external site.

containedInLinks to an external site.

SNOMED-CT provides a high level concept of Linkage concept under which there are many potential relationships, such as:

Cause ofLinks to an external site.

o

Due toLinks to an external site.

Contained inLinks to an external site.

Treated withLinks to an external site.

Brought on byLinks to an external site.

.

A common way to represent a knowledge graph is with

semantic triplesLinks to an external site.

(a kind of 3-tuple).

For this assignment you will create a 4-tuple where in addition to the subject, predicate, object of the semantic triple you will add a fourth concept of evidence.

Therefore, each relationship (edge) in the graph should be justified in terms of evidence. Evidence can be in terms of published literature (e.g. PubMed ID), practice guidelines from professional organizations, personal experience, etc. You should provide the most reliable evidence available. Evidence should be referenced with a URL. As with self-generated definitions, any personally generated evidence needs to be available at some URL (again, via GitHub, Google Docs, etc.).

Example

The following knowledge graph is recreated from

this conference paperLinks to an external site.

:

One of the relationships in the graph would be represented as follows (this is a YAML dictionary rather than a tuple):

– subject:

pelvic pain

predicate:

may be finding of disease

object: pelvic lipamatosis

To make this relationship satisfactory for this assignment, I need to provide reference definitions and evidence. For example:

– subject:

concept: pelvic pain

id:

http://purl.obolibrary.org/obo/HP_0034267

object:

concept: pelvic-lipmatosis

id:

http://purl.bioontology.org/ontology/SNOMEDCT/1908…

predicate:

concept: Sign or Symptom

id:

http://purl.bioontology.org/ontology/STY/T184

evidence:

–

https://www.ncbi.nlm.nih.gov/pmc/articles/pmid/381…

I have used SNOMED-CT to define the concepts, I’ve simplified “may be finding of disease” to simply “Sign or Symptom [of]” (the “of” is implicit in the arrow), and I’ve provided evidence for this relationship, in this case a single PubMed Central article. You could have multiple sources of evidence, but you should have at least one.

Provide with your knowledge graph a brief written description of your knowledge graph that addresses at the following three of the five components of a knowledge representation enumerated by Davis, Shrobe, and Szolovits:

1.Surrogate: What is the health problem this knowledge graph is a surrogate for?

2.Ontological commitments: What are the key concepts you are focusing on and what are you explicitly ignoring?

3.Fragmentary theory of intelligent reasoning: What kind of reasoning does your knowledge graph support?