# COVID-19 corpus of research articles annotated with biomedical entities

Posted on March 28, 2020 by David Campos ( ) 8 minute read

# TL;DR

Created a corpus of research articles related with COVID-19, automatically annotated with 10 biomedical entities of interest, namely Disorder, Species, Chemical or Drug, Gene or Protein, Enzyme, Anatomy, Biological Process, Molecular Function, Cellular Component, Pathway and microRNA.

The corpus is freely available and can be used to further research topics related with COVID-19, contributing to find insights towards a better understanding of the disease, in order to find effective drugs and reduce the pandemic impact.

Please follow the progress on Github, which already provides the CORD-19 corpus of full-text articles with more then 31 million biomedical annotations.

# Statistics

Overall corpus statistics:

• Number of abstracts: 17740
• Number of entity occurrences: 683349
• Number of unique entities: 29423

Number of annotations per entity type:

Entity # Occurrences # Unique
Disorder 183528 4477
Species 128356 2170
Chemical or Drug 70619 2768
Gene and Protein 51114 15025
Enzyme 7892 282
Anatomy 106401 2369
Biological Process 74286 1561
Molecular Function 15089 383
Cellular Component 39451 263
Pathway 6587 97
microRNA 26 28

# Structure

Corpus file corpus/pubmed_YYYYMMDD.zip contains the following folders:

• json: article in JSON format from Pubmed;
• raw: article with text only;
• annotations: annotations in A1 format.

On each folder you can find one file per article, with the Pubmed ID on its name.

# Articles

To collect articles related with COVID-19 from Pubmed, the following query was applied:


1
("2000"[Date - Publication] : "3000"[Date - Publication]) AND ((COVID-19) OR (Coronavirus) OR (Corona virus) OR (2019-nCoV) OR (SARS-CoV) OR (MERS-CoV) OR (Severe Acute Respiratory Syndrome) OR (Middle East Respiratory Syndrome) OR (2019 novel coronavirus disease[MeSH Terms]) OR (2019 novel coronavirus infection[MeSH Terms]) OR (2019-nCoV disease[MeSH Terms]) OR (2019-nCoV infection[MeSH Terms]) OR (coronavirus disease 2019[MeSH Terms]) OR (coronavirus disease-19[MeSH Terms]))


To collect the articles in JSON and then extract the text in raw format:


1
2
python scripts/pubmed/pubmed.py
python scripts/pubmed/json2raw.py


Please not that sentences in other languages than english are currently being discarded.

# Resources

The following resources were applied to annotated each entity type:

For more details please check the article. Unfortunately dictionaries could not be shared for download, due to UMLS usage license. Nevertheless, keep in mind that Disorder and Species entities were extended to include COVID-19 entities of interest.

# Annotation

Neji is the tool used for NER (Named Entity Recognition) and normalization, which is optimized for biomedical scientific articles and provides an easy to use CLI. For more details please check the article.

The annotation script is available at scripts/pubmed/annotate.sh.

# Visualization

brat is used to visualize the annotations in the articles. Find below the instructions to run the tool, create corpus for brat and visualize annotations.

##### Install and run brat

1
2
3
4
5
cd tools
unzip brat-1.3.zip
cd brat-1.3
./install.sh -u
python standalone.py

##### Create corpus for visualization

1
2
./scripts/pubmed/brat.sh
ln -s corpus/pubmed/brat tools/brat-1.3/data/covid19-corpus

##### Visualize corpus

Go to http://localhost:8001/index.xhtml#/covid19-corpus/ and wait for the articles to load:

Figure: List of articles.

Double click in a document to visualize it:

Figure: Article with annotations visualization.

# Example

Find below an example article in the provided document formats: JSON, Raw and A1.

##### JSON

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
{
"pubmed_id": "32198088",
"title": "Transmission potential and severity of COVID-19 in South Korea.",
"abstract": "Since the first case of 2019 novel coronavirus (COVID-19) identified on Jan 20, 2020 in South Korea, the number of cases rapidly increased, resulting in 6,284 cases including 42 deaths as of March 6, 2020. To examine the growth rate of the outbreak, we aimed to present the first study to report the reproduction number of COVID-19 in South Korea.\nThe daily confirmed cases of COVID-19 in South Korea were extracted from publicly available sources. By using the empirical reporting delay distribution and simulating the generalized growth model, we estimated the effective reproduction number based on the discretized probability distribution of the generation interval.\nWe identified four major clusters and estimated the reproduction number at 1.5 (95% CI: 1.4-1.6). In addition, the intrinsic growth rate was estimated at 0.6 (95% CI: 0.6, 0.7) and the scaling of growth parameter was estimated at 0.8 (95% CI: 0.7, 0.8), indicating sub-exponential growth dynamics of COVID-19. The crude case fatality rate is higher among males (1.1%) compared to females (0.4%) and increases with older age.\nOur results indicate early sustained transmission of COVID-19 in South Korea and support the implementation of social distancing measures to rapidly control the outbreak.",
"keywords": [
"COVID-19",
"Korea",
"coronavirus",
"reproduction number"
],
"journal": "International journal of infectious diseases : IJID : official publication of the International Society for Infectious Diseases",
"publication_date": "2020-03-22",
"authors": [
{
"lastname": "Shim",
"firstname": "Eunha",
"initials": "E",
"affiliation": "Department of Mathematics, Soongsil University, 369 Sangdoro, Dongjak-Gu, Seoul, 06978 Republic of Korea. Electronic address: alicia@ssu.ac.kr."
},
{
"lastname": "Tariq",
"firstname": "Amna",
"initials": "A",
"affiliation": "Department of Population Health Sciences, School of Public Health, Georgia State University, Atlanta, GA, USA. Electronic address: atariq1@student.gsu.edu."
},
{
"lastname": "Choi",
"firstname": "Wongyeong",
"initials": "W",
"affiliation": "Department of Mathematics, Soongsil University, 369 Sangdoro, Dongjak-Gu, Seoul, 06978 Republic of Korea. Electronic address: chok10004@soongsil.ac.kr."
},
{
"lastname": "Lee",
"firstname": "Yiseul",
"initials": "Y",
"affiliation": "Department of Population Health Sciences, School of Public Health, Georgia State University, Atlanta, GA, USA. Electronic address: ylee97@student.gsu.edu."
},
{
"lastname": "Chowell",
"firstname": "Gerardo",
"initials": "G",
"affiliation": "Department of Population Health Sciences, School of Public Health, Georgia State University, Atlanta, GA, USA. Electronic address: gchowell@gsu.edu."
}
],
"methods": null,
"conclusions": null,
"results": "We identified four major clusters and estimated the reproduction number at 1.5 (95% CI: 1.4-1.6). In addition, the intrinsic growth rate was estimated at 0.6 (95% CI: 0.6, 0.7) and the scaling of growth parameter was estimated at 0.8 (95% CI: 0.7, 0.8), indicating sub-exponential growth dynamics of COVID-19. The crude case fatality rate is higher among males (1.1%) compared to females (0.4%) and increases with older age.",
"doi": "10.1016/j.ijid.2020.03.031",
"xml": null
}

##### Raw

1
2
3
4
5
6
7
8
TITLE:
Transmission potential and severity of COVID-19 in South Korea.

ABSTRACT:
Since the first case of 2019 novel coronavirus (COVID-19) identified on Jan 20, 2020 in South Korea, the number of cases rapidly increased, resulting in 6,284 cases including 42 deaths as of March 6, 2020. To examine the growth rate of the outbreak, we aimed to present the first study to report the reproduction number of COVID-19 in South Korea.
The daily confirmed cases of COVID-19 in South Korea were extracted from publicly available sources. By using the empirical reporting delay distribution and simulating the generalized growth model, we estimated the effective reproduction number based on the discretized probability distribution of the generation interval.
We identified four major clusters and estimated the reproduction number at 1.5 (95% CI: 1.4-1.6). In addition, the intrinsic growth rate was estimated at 0.6 (95% CI: 0.6, 0.7) and the scaling of growth parameter was estimated at 0.8 (95% CI: 0.7, 0.8), indicating sub-exponential growth dynamics of COVID-19. The crude case fatality rate is higher among males (1.1%) compared to females (0.4%) and increases with older age.
Our results indicate early sustained transmission of COVID-19 in South Korea and support the implementation of social distancing measures to rapidly control the outbreak.

##### Annotations

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
T0	DISO 46 54	COVID-19
N0	Reference T0 UMLS:::DISO	COVID-19
T1	SPEC 106 128	2019 novel coronavirus
N1	Reference T1 NCBI:2697049:T001:SPEC	2019 novel coronavirus
T2	DISO 130 138	COVID-19
N2	Reference T2 UMLS:::DISO	COVID-19
T3	PROC 260 266	deaths
N3	Reference T3 GO:0016265::PROC	deaths
T4	PROC 303 309	growth
N4	Reference T4 UMLS:C1621966:T042:PROC	growth
N5	Reference T4 GO:0040007::PROC	growth
T5	PROC 382 394	reproduction
N6	Reference T5 GO:0000003::PROC	reproduction
T6	DISO 405 413	COVID-19
N7	Reference T6 UMLS:::DISO	COVID-19
T7	DISO 459 467	COVID-19
N8	Reference T7 UMLS:::DISO	COVID-19
T8	PROC 602 620	generalized growth
N9	Reference T8 GO:0040007::PROC	generalized growth
T9	PROC 614 620	growth
N10	Reference T9 UMLS:C1621966:T042:PROC	growth
T10	PROC 655 667	reproduction
N11	Reference T10 GO:0000003::PROC	reproduction
T11	CHED 778 786	clusters
N12	Reference T11 CHEBI:33731:T103:CHED	clusters
T12	PROC 805 817	reproduction
N13	Reference T12 GO:0000003::PROC	reproduction
T13	PROC 878 884	growth
N14	Reference T13 UMLS:C1621966:T042:PROC	growth
N15	Reference T13 GO:0040007::PROC	growth
T14	PROC 949 955	growth
N16	Reference T14 UMLS:C1621966:T042:PROC	growth
N17	Reference T14 GO:0040007::PROC	growth
T15	PROC 1034 1040	growth
N18	Reference T15 UMLS:C1621966:T042:PROC	growth
N19	Reference T15 GO:0040007::PROC	growth
T16	DISO 1053 1061	COVID-19
N20	Reference T16 UMLS:::DISO	COVID-19
T17	DISO 1231 1239	COVID-19
N21	Reference T17 UMLS:::DISO	COVID-19


# Next steps

Possible next steps to improve the COVID-19 corpus:

• Annotate “methods”, “results” and “conclusions” sections from JSON files;
• Further optimize resources to target entities related with COVID-19;
• Include additional entities of relevance;
• Annotate PMC and Elsevier full text articles;
• Collect co-occurrences to understand which entities might be related more often;