CORD 19 Publication Processing

CORD-19 Processing Details

COVID-19 Open Research Dataset (CORD-19)

Version 1 (at the time of Hackathon) : Metamap 2018 was used. 7 semantic types were extracted.

Version 2 : Metamap 2020 was used to extract entities and Semrep 2016 was used to extract semantic relations from the publications. The annotations included many subject and object arguments incorrectly normalised to CUI in UMLS. Eg: Mentions of mask in a covid related paper were linked to a Gene in UMLS. Phrase: mask Meta Mapping (1000): 1000 C1538279:MASK (ANKHD1 gene) [Gene or Genome]

Mentions of probability values in the text incorrectly identified as medical entities:
Phrase: (P < .001

Meta Mapping (861): 861 C0369773:PNOS (P Blood group antibodies) [Amino Acid, Peptide, or Protein,Immunologic Factor]

Version 3: NIH has made available Semrep tool results on CORD-19 dataset. We use this results to extract both the entities and relations from the publications.

SemRep Options Used for Processing: semrep -A -N -n -S -F -Z 2020AA A: Anaphora resolution N: use_generic_domain_extension n: use_generic_domain_modification S: generic_processing F: full_fielded_output Z 2020AA: use 2020AA data

We used Semrep processed CORD-19 release dated 02/15/2021. Number of publications processed: 97,441

---

Processing the Semrep Results

For Entities:

To prune the annotations for our research goal, we include biomedical entities from the publications that belong to only one or more of the following UMLS Semantic Types: CHEM|Chemicals & Drugs|T116|Amino Acid, Peptide, or Protein|aapp CHEM|Chemicals & Drugs|T195|Antibiotic|antb CHEM|Chemicals & Drugs|T123|Biologically Active Substance|bacs CHEM|Chemicals & Drugs|T103|Chemical|chem CHEM|Chemicals & Drugs|T200|Clinical Drug|clnd CHEM|Chemicals & Drugs|T126|Enzyme|enzy CHEM|Chemicals & Drugs|T197|Inorganic Chemical|inch CHEM|Chemicals & Drugs|T114|Nucleic Acid, Nucleoside, or Nucleotide|nnon CHEM|Chemicals & Drugs|T109|Organic Chemical|orch CHEM|Chemicals & Drugs|T121|Pharmacologic Substance|phsu DISO|Disorders|T020|Acquired Abnormality|acab DISO|Disorders|T049|Cell or Molecular Dysfunction|comd DISO|Disorders|T047|Disease or Syndrome|dysn DISO|Disorders|T037|Injury or Poisoning|inpo DISO|Disorders|T191|Neoplastic Process|neop DISO|Disorders|T046|Pathologic Function|patf DISO|Disorders|T048|Mental or Behavioral Dysfunction|mobd DISO|Disorders|T184|Sign or Symptom|sosy GENE|Genes & Molecular Sequences|T087|Amino Acid Sequence|amas GENE|Genes & Molecular Sequences|T028|Gene or Genome|gngm GENE|Genes & Molecular Sequences|T088|Carbohydrate Sequence|crbs GENE|Genes & Molecular Sequences|T085|Molecular Sequence|mosq GENE|Genes & Molecular Sequences|T086|Nucleotide Sequence|nusq LIVB|Living Beings|T005|Virus|virs

**For Semantic Relations: ** We excluded a subset of predicate types that were not helpful to understand toxicities in COVID related treatments such as PART _ OF and PROCESS _ OF. The predicate types we used are:

AFFECTS , ASSOCIATED _ WITH , AUGMENTS , CAUSES , COEXISTS _ WITH , COMPLICATES , DISRUPTS , INHIBITS , INTERACTS _ WITH , MANIFESTATION _ OF , PREDISPOSES , PREVENTS , PRODUCES , STIMULATES , USES, PRECEDES and TREATS .

We also included only those relations in which the subject or the object belongs to one or more of the semantic types mentioned above for entities.