Gender Bias in Bioinformatics

This repo contains the scripts and data used to examine gender bias in Bioinformatics. By taking five representative journals in the field—Oxford Bioinformatics, Plos Computational Biology, Nucleic Acids Research, BMC Bioinformatics, and BMC Genomics—,we conduct a large-scale analysis of the role of female researchers in Bioinformatics from 2005 to 2017. We chose to start in 2005 because this was the starting year of Plos Computational Biology.

Data Collection

Data were obtained from Scopus, one of today's most complete repository of scientific manuscripts. The data was collected on August 22nd and 23rd, 2019, so the obtained articles correspond to the information available on Scopus at that moment. Below, the steps were performed to get the data.

• Query the search engine of Scopus using the next string through Advanced Search to extract articles published between 2005 and 2017 from above-mentioned journals.

ISSN ( 'JOURNAL_ISSN' ) AND ( LIMIT-TO ( DOCTYPE,"ar" ) OR LIMIT-TO ( DOCTYPE,"cp" ) ) AND ( LIMIT-TO ( PUBYEAR , 2017 ) OR LIMIT-TO ( PUBYEAR , 2016 ) OR LIMIT-TO ( PUBYEAR , 2015 ) OR LIMIT-TO ( PUBYEAR , 2014 ) OR LIMIT-TO ( PUBYEAR , 2013 ) OR LIMIT-TO ( PUBYEAR , 2012 ) OR LIMIT-TO ( PUBYEAR , 2011 ) OR LIMIT-TO ( PUBYEAR , 2010 ) OR LIMIT-TO ( PUBYEAR , 2009 ) OR LIMIT-TO ( PUBYEAR , 2008 ) OR LIMIT-TO ( PUBYEAR , 2007 ) OR LIMIT-TO ( PUBYEAR , 2006 ) OR LIMIT-TO ( PUBYEAR , 2005 ) )

Journal	ISSN	Papers Extracted
Oxford Bioinformatics	1460-2059	8,546
Plos Computational Biology	1553-734X	5,132
Nucleic Acids Research	1362-4962	15,670
BMC Bioinformatics	1471-2105	7,879
BMC Genomics	1471-2164	10,200
Total		47,427

• Use the function Export to download the data about the articles. CSV was chosen as the export method and all of the information available per article (citation, bibliographical, abstract, funding, etc.) was asked to export. Here it is important to mention that Scopus limits to 2,000 the number of records that can be exported at a time, so in some situations, the range of years (2005-2017) was split into several searches to comply with this restriction.

The raw data downloaded can be found in CSV files located in data/raw/full. The data/raw/summary directory contains files with only citation information about the articles. In total, information of 47,427 papers and their corresponding authors were collected through the described method. The table above shows the number of papers per journal extracted from Scopus.

Getting Started

1. Download and install Python >= 3.4.4;
2. Download and install MongoDB community version. Instructions on how to install it can be found here;
3. Set up a Mongodb database and create two collections, namely bioinfo_papers and bioinfo_authors;
4. Clone the repository git clone https://github.com/ParticipaPY/politic-bots.git;
5. Get into the directory of the repository cd politic-bots;
6. Create a virtual environment by running virtualenv env;
7. Activate the virtual environment by executing source env/bin/activate;
8. Inside the directory of the repository install the project dependencies by running pip install -r requirements.txt;
9. Set the database information inside the dictionary mongo in config.json;
10. Get a key to operate the API of PubMed by following the instructions here
11. Set the obtained API key and email address inside the dictionary pubmed in config.json;
12. Run run.py to go through all of the data pre-processing, loading, and exporting tasks. Three CSV files result from the execution of run.py, they are stored under the data directory.

The following sections explain in details each of the pre-processing, loading, and exporting tasks.

Data Pre-Processing

From run.py execute the function combine_csv_files in data_wrangler.py to combine files in data/raw/full into one CSV file per journal. The resulting files are stored in data/processed (you might need to create the folder processed inside data before running the function)

Data Loading

From run.py execute the function load_data_from_files_into_db in data_loader.py to load the data in data/raw/summary and data/processed into the database. Information about papers is stored in bioinfo_papers while information on papers' authors is recorded in bioinfo_authors. This function takes a while to complete in part because it connects to DOI resolution to extract links of the papers—Scopus does not provide links to the papers. The completion time can be sped up by commenting the line #162 in data_loader.py.

Duplicated records (194) and entries without DOI (401) are not stored. In total, 46,832 records are stored in the database. The distribution of duplicated articles and articles without DOI per journal is shown in the next table.

Journal	Duplicates	Missing DOIs
Oxford Bioinformatics	79	1
Plos Computational Biology	0	11
Nucleic Acids Research	12	169
BMC Bioinformatics	76	100
BMC Genomics	27	120
Total	194	401

Data Processing

Scopus does not provide the full name of authors—only the initial of the first (and middle) name and the last name. However, the PubMed identifier of the articles is provided by Scopus. We use the PubMed Id of the articles to hit the API of PubMed and get information about the papers' authors, including their full names. To complete the name of authors, execute from run.py the function get_paper_author_names_from_pubmed in data_extractor.py. The function takes a while to complete.

Gender Identification. The API NamSor is hit to infer the authors' gender from the author's name. In case, NamSor fails to identify the gender, the python package gender-guesser is used to find out the gender of authors. Information on how NamSor works can be at its website.

Through this process, we find that 266 articles (0.6%) are not in PubMed, so the information about their authors cannot be obtained from this source. For different reasons, we cannot get information about 12 articles that have the PubMed identifier. Ten of them are proceedings of conferences, 1 is a PDF with the names of the editorial board of the journal, and 1 does not have author list.

We cannot get information on 2,626 authors (0.18%). In some cases, Scopus does not provide the last name specifying the situation with an empty string or with the text [No author name available]. In some other cases, there are inconsistencies between the list of authors provided by Scopus and the list of authors obtained from PubMed. This is the case of articles in which organizations appear as part of the author list. Here, PubMed mentions the organization's name while Scopus present the name of the organization's members that authored the article. We also found PubMed registries with the first and last name inverted. These registries cannot be automatically matched.

In total, the gender of 27,706 authors (19%) cannot be detected by not either of the two gender identification services. In 10% of the cases, the gender cannot be identified because we cannot get the author name from PubMed. For the rest, we found that the identification services have problems with Asian names.

Data Export

Before running the analyses, data are exported to tabular format and saved into CSV files. Data about papers can be exported to a CSV by running the function export_db_into_file included in data_exporter.py. The function receives as parameters, the name of the CSV file, the database from where to extract the data, and the fields to be extracted. The following code snippet shows an example of how to export data about papers into the CSV file paper.csv, which is saved in the directory data.

from data_exporter import export_db_into_file
from db_manager import DBManager
from utils import get_db_name
    
db_papers = DBManager('bioinfo_papers', db_name=get_db_name())
fields_to_export = ['title', 'DOI', 'year', 'source', 'citations', 
                    'edamCategory', 'link', 'authors', 
                    'gender_last_author']
export_db_into_file('papers.csv', db_papers, fields_to_export)

In the same manner, information about authors can be exported to a CSV file. The function export_author_papers in data_exported.py creates a CSV that registers the cartesian product between papers and authors. The resulting file is saved into the directory data.

Gender Bias Analysis

The CSV files resulting from the exporting task (i.e., data/papers.csv, data/authors.csv, and data/papers_authors.csv), are used to conduct the gender bias analyses. The analysis scripts are contained in the notebook analysis/gender_bias_analysis.ipynb.

Technologies

1. Python 3.4
2. MongoDB Community Edition—used as data storage repository
3. Selenium WebDriver—used to resolve papers' DOIs
4. Biopython—PubMed API client
5. Jupyter Notebook—data exploration and analysis

Issues

Please use Github's issue tracker to report issues and suggestions.

Contributors

Jorge Saldivar, Fabio Curi, Nataly Buslón, María José Rementería, and Alfonso Valencia