license: cc-by-2.0
task_categories:
- text-classification
language:
- en
tags:
- biology
pretty_name: RNAcentral Litscan ncRNA Related Abstracts
size_categories:
- 1K<n<10K
RNAcentral Litscan ncRNA Related Abstracts
This is a dataset of abstracts that have been labelled relevant to ncRNA or not. Relevant articles have the label 1, irrelevant have the label 0
How was it made
The dataset was built in five parts:
- ncRNA related articles from tarbase
- ncRNA related articles from Rfam
- ncRNA related articles from GO curation
- Manually annotated ncRNA related articles
- Non ncRNA related articles
TarBase
TarBase is a database of miRNAs and their targets. This data is curated from the literature, so by definition, the papers they cite for the interactions are relevant for ncRNA. This is somewhat limited to miRNA though
Rfam
Rfam is the ncRNA families database, and has families for >4000 RNAs. They are mostly miRNAs, but also cover a lot of other types, and the non-miRNA part is expanding. Rfam families are usually built from an alignmentthat is published, so the papers mentioned by Rfam families will be relevant to ncRNA
GO
GO is the Gene Ontology, a knowledge base of function in biology. A subset of annotations in GO are manually curated by human curators reading papers, so those papers curated with ncRNA relevant terms will definitely be relevant to ncRNA. A large fraction of the GO curation is about miRNA-mRNA interactions, but there is a bit of other stuff in there as well; reflecting the type of analyses done in low throughput experiments on ncRNA
Manually annotated articles
A set of ~400 abstracts were manually assessed for relevance to ncRNA by members of the RNAcentral & Rfam teams. These were drawn from a EuropePMC query designed to pull in as many potentially relevant articles as possible. These should cover all RNA types, but the set is relatively small
Non-ncRNA related articles
To provide our negative set, we used this EuropePMC search to retrieve ~3500 articles:
f'/search?query=(IN_EPMC:Y AND OPEN_ACCESS:Y AND NOT SRC:PPR AND NOT "rna" ' \
f'AND NOT "mrna" AND NOT "ncrna" AND NOT "lncrna" AND NOT "rrna" AND NOT "sncrna" AND NOT "mirna") ' \
f'&sort_cited:y&pageSize=500&cursorMark={page}&format=json'
This is the opposite query to the one used in production for LitScan, so should be pulling in totally irrelevant articles.
Limitations
- Most of the positive set come from annotations or other linkages with miRNAs. Therefore, we kind of expect this dataset to produce a model that is good at finding miRNA related articles, but it may struggle when filtering articles relevant for other types (e.g. snoRNA) that are under-represented
- The negative set may be too easy to separate from the positive, since they are deliberately very irrelevant