In ropensci/tidypmc: Parse Full Text XML Documents from PubMed Central

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "# "
)

tidypmc

The Open Access subset of Pubmed Central (PMC) includes 2.5 million articles from biomedical and life sciences journals. The full text XML files are freely available for text mining from the REST service or FTP site but can be challenging to parse. For example, section tags are nested to arbitrary depths, formulas and tables may return incomprehensible text blobs and superscripted references are pasted at the end of words. The functions in the tidypmc package are intended to return readable text and maintain the document structure, so gene names and other terms can be associated with specific sections, paragraphs, sentences or table rows.

Installation

Use remotes to install the package.

remotes::install_github("ropensci/tidypmc")

Load XML

Download a single XML document like PMC2231364 from the REST service using the pmc_xml function.

options(width=100)
library(tidypmc)
library(tidyverse)
doc <- pmc_xml("PMC2231364")
doc

The europepmc package includes additional functions to search PMC and download full text. Be sure to include the OPEN_ACCESS field in the search since these are the only articles with full text XML available.

options(width=100)
library(europepmc)
yp <- epmc_search("title:(Yersinia pestis virulence) OPEN_ACCESS:Y")
select(yp, pmcid, pubYear, title) %>%
  print(n=5)

Save all r nrow(yp) results to a list of XML documents using the epmc_ftxt or pmc_xml function.

docs <- map(yp$pmcid, epmc_ftxt)

See the PMC FTP vignette for details on parsing the large XML files on the FTP site with 10,000 articles each.

Parse XML

The package includes five functions to parse the xml_document.

|R function |Description | |:--------------|:--------------------------------------------------------------------------| |pmc_text |Split section paragraphs into sentences with full path to subsection titles| |pmc_caption |Split figure, table and supplementary material captions into sentences | |pmc_table |Convert table nodes into a list of tibbles | |pmc_reference|Format references cited into a tibble | |pmc_metadata |List journal and article metadata in front node |

The pmc_text function uses the tokenizers package to split section paragraphs into sentences. The function also removes any tables, figures or formulas that are nested within paragraph tags, replaces superscripted references with brackets, adds carets and underscores to other superscripts and subscripts and includes the full path to the subsection title.

options(width=110)
txt <- pmc_text(doc)
txt
count(txt, section, sort=TRUE)

Load the tidytext package for further text processing.

options(width=110)
library(tidytext)
x1 <- unnest_tokens(txt, word, text) %>%
  anti_join(stop_words) %>%
  filter(!word %in% 1:100)
filter(x1, str_detect(section, "^Results"))
filter(x1, str_detect(section, "^Results")) %>%
  count(word, sort = TRUE)

The pmc_table function formats tables by collapsing multiline headers, expanding rowspan and colspan attributes and adding subheadings into a new column.

options(width=110)
tbls <- pmc_table(doc)
map_int(tbls, nrow)
tbls[[1]]

Use collapse_rows to join column names and cell values in a semi-colon delimited string (and then search using functions in the next section).

options(width=110)
collapse_rows(tbls, na.string="-")

The other three pmc functions are described in the package vignette.

Searching text

There are a few functions to search within the pmc_text or collapsed pmc_table output. separate_text uses the stringr package to extract any regular expression or vector of words.

options(width=110)
separate_text(txt, "[ATCGN]{5,}")

A few wrappers search pre-defined patterns and add an extra step to expand matched ranges. separate_refs matches references within brackets using \\[[0-9, -]+\\] and expands ranges like [7-9].

options(width=110)
separate_refs(txt)

separate_genes will find microbial genes like tauD (with a capitalized 4th letter) and expand operons like tauABCD into four genes. separate_tags will find and expand locus tag ranges below.

options(width=110)
collapse_rows(tbls, na="-") %>%
  separate_tags("YPO") %>%
  filter(id == "YPO1855")

See the vignette for more details including code to parse XML documents using the xml2 package. The PMC FTP vignette has details on parsing XML files at the Europe PMC FTP site.

Community Guidelines

This project is released with a Contributor Code of Conduct. By participating in this project you agree to abide by its terms. Feedback, bug reports, and feature requests are welcome here.

ropensci/tidypmc documentation built on Dec. 14, 2019, 11:42 p.m.