Name: bioentities
Owner: WikiPathways
Description: Namespace encoding hierarchical relationships between proteins, protein families, and protein complexes.
Forked from: johnbachman/bioentities
Created: 2018-01-11 21:05:38.0
Updated: 2018-04-13 01:51:05.0
Pushed: 2018-04-13 01:51:03.0
Homepage: null
Size: 560
Language: Python
GitHub Committers
User | Most Recent Commit | # Commits |
---|
Other Committers
User | Most Recent Commit | # Commits |
---|
Bioentities is a collection of resources for grounding biological entities from text and describing their hierarchical relationships. Resources were developed by manual curation for use by natural language processing and biological modeling teams in the DARPA Big Mechanism and Communicating with Computers programs. The repository contains the following files:
lies and protein complexes. For example, ```PIK3CA isa PIK3C```, where
C represents the class of catalytic subunits of PI3K; and ```PIK3C partof
```, where PI3K represents a named complex consisting of a catalytic and
latory subunit.
Bioentities namespace.
ntities namespace.
ogical databases.
he cross-references among the various files.
Bioentities contains resources for defining the relationships between
genes/proteins and their membership in families and named complexes. Entities
defined within the Bioentities namespace are listed in the `entities.csv
`
file. Cross-referencing the entries among the various files maintains
consistency and prevents errors.
Relationships are defined in `relations.csv
` as a triples using two
relationships:
These two relationships can be combined to capture complex hierarchical relationships, including sub-families (families within families) and complexes consisting of families of related subunits (e.g., PI3K, NF-kB).
The `relations.csv
file consists of five columns: (1) the namespace for
the subject (e.g., ``
HGNC` for gene names,
UP``
for Uniprot, or `BE
for the Bioentities namespace), (2) the identifier for the subject, (3) the
relationship (``
isa` or
partof``
), (4) the namespace for the object, and
(5) the identifier for the object.
The `equivalences.csv
file consists of three columns (1) the namespace of
an outsite entity (e.g. ``
BEL`,
PFAM``
),
(2) the identifier of the outside entity in the namespace given in the
first column, and (3) the equivalent entity in the `BE
` namespace.
Using mechanisms extracted from text mining to explain biological datasets requires that the entities in text are correctly grounded to the canonical names and IDs of genes, proteins, and chemicals. The problem is that simple lookups based on string matching often fail, particularly for protein families and named complexes, which appear frequently in text but lack corresponding entries in databases.
The grounding map addresses this by providing explicit grounding for frequently encountered entities in the biological literature. The text strings were drawn from a corpus of roughly 32,000 papers focused on growth factor signaling in cancer.
Entities are grounded to the following databases:
Genes/proteins: Uniprot
Protein families and named complexes: grounded to entities defined within
the Bioentities repository in the `entities.csv
and ``
relations.csv```
files, and to identifiers in PFAM
and Interpro when possible.
Note: Some text strings in the map have no grounding. This was originally used to identify entities that represent parsing errors and that should not be included in downstream output. For example, “MAP” a degenerate extraction that could signify many entities, including MAP kinase, MAP kinase inhibitor, MAP kinase kinase, etc. However, these empty entries could be used differently depending on the downstream application.
The file `gene_prefixes.csv
` enumerates prefixes and suffixes frequently
appended to named entities. Some of these represent subtleties of experimental
context (for example, that a protein of interest was tagged with a fluorescent
protein in an experiment) that can safely be ignored when determining the logic
of a sentence. However, others carry essential meaning: for example, a sentence
describing the effect of 'AKT shRNA' on a downstream target has the opposite
meaning of a sentence involving 'AKT', because 'AKT shRNA' represents
inhibition of AKT by genetic silencing.
The patterns included in this file were found by manually reviewing 70,000 named entities extracted by the REACH parser from a corpus of roughly 32,000 papers focused on growth factor signaling.
**Important note: the prefixes/suffixes may be applied additively, for example
The file contains three columns:
`mEGFP-{Gene name}
, where ``
{Gene name}``` represents a protein/gene name.The category of the prefix/suffix determines whether it can be stripped off with minimal effect on the meaning, or whether it carries meaning that needs to be incorporated by a parser. The categories are as follows:
rally be ignored.**
gene. **In most use cases can be ignored.**
might designate that an entity is a "protein", a "protease",
nscription factor", etc. **In most use cases can be ignored.**
case of ```{Gene name} mRNA```, the entity **must be explicitly grounded
n mRNA.**
fication, cellular localization, etc. **Must be captured by the
er.**
bition of the protein, that is, a loss-of-function experiment. Have the
ct of switching the polarity of the extracted mechanism. For example, the
ence "DUSP6 silencing leads to MAPK1 phosphorylation" indicates that DUSP6
hibits** MAPK1 phosphorylation. **Must be captured by the parser.**
Contributions are welcome! If making additions or revisions to the CSV files take care to handle quotations and newlines correctly. This allows diffs to be handled correctly so changes can be reviewed. Please submit updates via pull requests on Github.
The CSV files in the Bioentities repo are set up to be edited natively using Microsoft Excel. The CSV files in the repo have Windows line terminators ('\r\n'), and are not ragged (i.e., missing entries in a row are padded out with empty strings to reach the full width of the longest row).
To preserve correct newlines, take the following steps:
If saving from Excel (Windows or Mac OS X), save to the “Windows Comma Separated (.csv)” format.
If reading (or writing) the files using a Python script, use the following set of csv format parameters::
csvreader = csv.reader(f, delimiter=',', quotechar='“',
quoting=csv.QUOTE_MINIMAL, lineterminator='\r\n')
If editing the files on Linux, post-process files using `unix2dos
` or a
similar program.