Estimating the scale of biomedical data generation using text mining

Abstract

and Transplantation (DAIT) 5 * Corresponding authors: GR: gabriel.rosenfeld@nih.gov, DL: dawei.lin@nih.gov 6 7 One Sentence Summary: Application of a word-embedding model trained on the 8 methods sections of research articles allows for estimation of the production of 9 diverse biomedical data types using text mining. 10 available for use under a CC0 license.peer-reviewed) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made The copyright holder for this preprint (which was not . http://dx.doi.org/10.1101/182857 doi: bioRxiv preprint first posted online Sep. 1, 2017; 2 Abstract 11While the impact of biomedical research has traditionally been measured using 12 bibliographic metrics such as citation or journal impact factor, the data itself is an 13 output which can be directly measured to provide additional context about a 14 publication's impact. Data are a resource that can be repurposed and reused 15 providing dividends on the original investment used to support the primary work. 16Moreover, it is the cornerstone upon which a tested hypothesis is rejected or 17 accepted and specific scientific conclusions are reached. Understanding how and 18where it is being produced enhances the transparency and reproducibility of the 19 biomedical research enterprise. Most biomedical data are not directly deposited in 20 data repositories and are instead found in the publication within figures or 21 attachments making it hard to measure. We attempted to address this challenge by 22 using recent advances in word embedding to identify the technical and 23 methodological features of terms used in the free text of articles' methods sections. 24We created term usage signatures for five types of biomedical research data, which 25 were used in univariate clustering to correctly identify a large fraction of positive 26 control articles and a set of manually annotated articles where generation of data 27 types could be validated. The approach was then used to estimate the fraction of 28 PLOS articles generating each biomedical data type over time. Out of all PLOS 29 articles analyzed (n = 129,918), ~7%, 19%, 12%, 18%, and 6% generated flow 30 cytometry, immunoassay, genomic microarray, microscopy, and high-throughput 31 sequencing data.