As part of our efforts to help the community with the CORD-19 challenge, we have released several resources to help.
- Sentence level annotations for the COVID-19 Data version 54 (1st October 2020 release)
- Entity co-occurrence data within sentences
- Vocabularies used in this work describing coronavirus family and (specifically) COVID-19
The large files in this repo are uploaded using GitHub Large File Sotre (LSF). To pull this content you can use the command:
git lfs clone https://github.com/SciBiteLabs/CORD19.git
To cite this data or read more about the work undertaken please use our preprint publication: Reference ontology and database annotation of the COVID-19 Open Research Dataset (CORD-19) https://doi.org/10.1101/2020.10.04.325266
Sentence level annotations for the COVID-19 Open Research Dataset Challenge (version 54) using biomedical ontologies from SciBite
https://github.com/SciBiteLabs/CORD19/tree/master/annotated-CORD-19
In March 2020 in response to the COVID-19 pandemic, The White House, AI2, CZI, MSR, Georgetown and NIH released an open research dataset along with a call for action (https://www.whitehouse.gov/briefings-statements/call-action-tech-community-new-machine-readable-covid-19-dataset/) from the science and technology community. The call asks that artificial intelligence experts to develop new text and data mining techniques that can help the science community answer high-priority scientific questions related to COVID-19. The original datasets can be accessed at Kaggle: https://www.kaggle.com/allen-institute-for-ai/CORD-19-research-challenge The annotations are made against version 54 of this data released 1st October 2020.
Our ontology and SciBiteAI teams have used our software to produce sentence level annotations on this dataset to help those trying to identify biomedical entities within the text. The following have been identified
| Ontology Type | Total Hits | Unique Hits |
|---|---|---|
| SPECIES - NCBI Taxon | 9958967 | 3502 |
| GOONTOL - Gene Ontology terms | 6975073 | 10702 |
| INDICATION - MeSH | 12998639 | 6360 |
| SARSCOV - coronavirus strains and viral processes | 2629924 | 302 |
| COUNTRY - Standard country codes | 2078689 | 228 |
| HPO - human phenotypes | 3997204 | 5737 |
| GENE - HGNC | 4018135 | 15714 |
| CORONAPROT - proteins of COVID-19 | 795869 | 67 |
| DRUG - ChEMBL | 2513491 | 20134 |
Identfying biomedical entities within sentences can be incredibly useful in narrowing down the space of 29,000 documents that researchers or algorithms are required to look at. For instance, if a specific gene or phenotype is of interest, then this data can be easily used to find those documents which contain them. Building knowledge graphs can be enhanced by being able to perform sentence cooccurence across large documents and forming edges depending upon the types contained. Sentence level sentiment/topic analysis using machine learning methods might also be peformed using the entity types. A full list of tasks are available on Kaggle: https://www.kaggle.com/allen-institute-for-ai/CORD-19-research-challenge/tasks
https://github.com/SciBiteLabs/CORD19/tree/master/sentence-co-occurrence-CORD-19/
This data set represents a way to quickly look for connections between entities in the literature. It has the following columns:
document_id, sentence_id, entity_uris, entity_types_plus_ids
In most cases the URIs will link to external resources (such as ontologies) which may be amenable to scraping and often have their own APIs so you can navigate out into the broader life science datascape should you wish to do so.
As an example of how you can use this to rapidly hone in on points of interest, we filtered out some sentences which have: a GENE or DRUG hit, and a CVPROT hit, and a hit for coronaviruses (an entity of type INDICATION):
MERS-CoV first binds, via its S protein, to the receptor DPP4 on the target cell [16] , and then releases its RNA genome, through plasma or endosomal membrane fusion, into the target cell. To demonstrate that the colocalization of hel/N complexes with M at late times postinfection was due to translocation of existing hel/N complexes to sites of M accumulation rather than to de novo synthesis of hel or N at a new location, we analyzed the effect of the translation inhibitor cycloheximide on hel/N relocalization (Fig. 4) . A 100 g/ml concentration of cycloheximide has been previously determined to be sufficient for inhibition of gene 1 translation in coronavirus-infected cells (Kim et al., 1995; Perlman et al., 1987; Sawicki and Sawicki, 1986) . When infected cells were incubated with cycloheximide beginning at 5.5 h p.i. and continuing throughout the remainder of infection, extensive colocalization of hel or N with M was still observed at 12 h p.i. (Figs. 4A and 4B ). However, on the basis of only these results, we could not determine whether these amino acid substitutions affected receptorbinding affinity between human DPP4 receptor and MERS-CoV S protein.
https://github.com/SciBiteLabs/CORD19/tree/master/vocabularies-CORD-19
Two vocabularies have been released, both of which were used for this work; a vocab specific to COVID-19 and one more generally around coronavirus family. This includes common names and synonyms for the viruses, common genes and proteins investigated in these strains along with relevant gene ontology terms.
If you have any questions or feedback or are interested in additional data annotations you can get in touch with us by emailing: cvdata@scibite.com or by raising a ticket in this GitHub repo.
The SciBite data annotations and vocabs are released under GPL https://github.com/SciBiteLabs/CORD19/blob/master/LICENSE