sap218 / ocimido

An ontology for ocular immune-mediated inflammatory diseases

OcIMIDo

Ocular Immune-Mediated Inflammatory Diseases Ontology

• Documentation
• Wiki
• Bioportal
• Manuscript

Important information

• Issues - see for ontology suggestions, bug reporting, and future development information.
• License - see before use (you can use/edit OcIMIDo as long as you give appropriate credit).
• Changelog - see major changes of OcIMIDo throughout development.

Creators

• Samantha Pendleton Institute of Cancer and Genomic Sciences, University of Birmingham, UK
• Tasanee Braithwaite The Medical Eye Unit, St Thomas’ Hospital NHS Foundation Trust, London UK

Citing

@article{PENDLETON2021104542,
title = {Development and application of the ocular immune-mediated inflammatory diseases ontology enhanced with synonyms from online patient support forum conversation},
journal = {Computers in Biology and Medicine},
volume = {135},
pages = {104542},
year = {2021},
issn = {0010-4825},
doi = {https://doi.org/10.1016/j.compbiomed.2021.104542},
url = {https://www.sciencedirect.com/science/article/pii/S001048252100336X},
author = {Samantha C. Pendleton and Luke T. Slater and Andreas Karwath and Rose M. Gilbert and Nicola Davis and Konrad Pesudovs and Xiaoxuan Liu and Alastair K. Denniston and Georgios V. Gkoutos and Tasanee Braithwaite},
}

ocimido's Issues

Text Splitting

• splitting via threads - all comments (not users)
• each thread sorted into bins: short/medium/long
• extract 20% from each bin and that is the test data set: put somewhere and not touch
• training is 80% of each bin
• extract synonyms for ontology

Check Ontology

td-idf of terms, add synonyms from forum train set

after #1 calculate word frequencies wrt length of threads using td-idf. doing this iteratively (removing terms directly expressed in the ontology each time) will inform you on synonyms and new terms to add, until you have coverage of all relevant terms in the ontology.

looks like the td-idf isn't in nltk directly, but you can use it to calculate the measure: https://nlpforhackers.io/tf-idf/

It looks like scikit-learn does have the functionality directly though: https://www.bogotobogo.com/python/NLTK/tf_idf_with_scikit-learn_NLTK.php

test coverage of ontology annotation on ov forum

after #3, when there is coverage of all threads in training set by the terms in the ontology, we should evaluate how this extrapolates to the test set, by annotating this in turn

we will have to discuss the best way to do this exactly, perhaps seeing if we have a similar number of labels per thread on average, or evaluate whether there's a similar distribution of labels across the graph (which might also produce a nice graphic for the paper)

Change ontology base

currently the fully qualified IRIs are e.g. https://github.com/sap218/ocular-immune-mediated-inflammatory-disease-ontology/blob/master/ontology/ocimido.owl#OCIMIDO_00001

This is a bit long, but more importantly we should change it to a domain we have some control over (the above becomes problematic if github ever design to change their design layout, you want to change your username, transfer it to a github org, github goes out of business etc) - I would recommend:

http://www.bham.ac.uk/OCIMIDO/

We should be able to open a ticket with IT services to get that URL redirected to here. The same goes for the ontologyIRI, as well as just the base defined in the ontology