sci.AI allows to explain two semantic types to the machines:
1. Meaning of the terms.
2. Interactions between biomedical objects in the text.
It is achieved by adding metadata layer to the plain text, report or dataset, whatever you process with sci.AI. This data layer consists of annotations added next to the text chunks.

Biomedical terms and relationships labeling in sci.AI Preprint

sci.AI Version: v.0.39
Document Version: v.0.81
Date: 30.10.2017
Authors & Editors: Roman Gurinovich, Alexander Pashuk, Vasili Puntus


Introduction

sci.AI allows to explain two semantic types to the machines:

  1. Meaning of the terms.
  2. Interactions between biomedical objects in the text.

It is achieved by adding metadata layer to the plain text, report or dataset, whatever you process with sci.AI. This data layer consists of annotations added next to the text chunks.

1. Process

  1. Import plain text to sci.AI.
  2. Validate terms and relationships labeled by sci.AI NLP Engine.
  3. Export machine-readable version of the text / paper / dataset in JATS .xml, .html with RDFa, RDF/XML etc.

Importing and exporting are described in separate documentation. Here we focus on validation of the automatic labeling and manual addition of the new labels.

2. Concepts Labeling

2.1. Confirm term-to-concept link

  1. Click on highlighted term in the text. All candidate concepts from the external ontologies will appear in sidebar on the left side of the screen.
  2. For each ontology, scroll to find correct concept.

  1. Click on “Correct Object”. It will hide all the rest candidates for the ontology automatically. Notice. It is very valuable to repeat for each proposed ontology and even add custom so that machine-readable output will contain term-ontology1-ontology2- … - ontology N links for the given context. It is a significant contribution to the global knowledge graph.

  1. Click on “Submit” when only correct concepts remain in sidebar. You can use “Submit and mark all the same”

2.2. Reject concept candidate as not valid for the term in this context

  1. Select term as in previous case.
  2. Scroll through the variants.
  3. Click on “Not this Object” for each non valid candidate.
  4. Click on “Submit” when all is done.

Note, that 2.1. And 2.2. are complementary. When you’ve found correct concept, it is not needed to scroll further: all the rest will be marked as non valid.

Only validated concepts will appear in final export file and knowledge graph. Skiping validation is identical to marking as "Not this object"

2.3. Add Custom Ontology Label

  1. Select term as in previous case.
  2. Add any ontology name or select from autosuggested. Then add ID of the concept in this ontology. You can add as many ontology+ID records as needed.
  3. Click on “Submit” when all is done.

2.4. Rejecting all labels for the term

  1. Click on “Not a Bio Object”. You can remove all the same selections in the text by clicking “Delete all the same”.

2.5. Labeling term that was not labeled by Engine

  1. Highlight the term or phrase with pointer. Engine will try to semanticize it again.
  2. Do any necessary step of validation or rejection. See, 2.1 - 2.4.

3. Authors labeling

Author validation is done by authentication via ORCID.

4. Labeling facts and relationships

At the moment, sci.AI supports labeling relationships, that are expressed in triple format. It is an experimental functionality and requires copying subject, predicate and object to the correspondent fields.

4.1. Adding relationship

  1. Click “Add Select sentence with fact”

  1. Copy and paste whole container sentence, subject, predicate and object.

  1. Submit.

4.2. Deleting relationship

  1. Select fact.
  2. Delete fact.

4.3. Confirming fact

  1. Click on underlined sentence.
  2. Press “Confirm the fact” if it is correct and contributes to the key knowledge of the paper.

5. Tables and Datasets Labeling

5.1. Semantic Labeling of the Datasets

Follow "key-value" like structure to allow semantic matching between phrase in the text and value in the table. Key in this case is URI of the object in external ontology.

As of v0.39 datasets need to be structured and labeled manually according to the recommendation above. Read more about semantic data in the literature [1].

5.2. Semantic Labeling of the Tables in the Text

Labeling of the tables in the text via sci.AI Preprint is done in the same way as concept labeling.

As of v0.39 it doesn't restructure the tables. Create it in SQL-like structure to allow future semantic matching of the values.

References

  1. Cheung K-H, Samwald M, Auerbach RK, Gerstein MB. Structured digital tables on the Semantic Web: toward a structured digital literature. Molecular Systems Biology. 2010;6:403. doi:10.1038/msb.2010.45.

Cookie Settings