Dataset

The Pubmed database and Thomson ISI are the sources of data for the trends shown. Most articles in PubMed have a list of ‘MeSH terms’ and ‘substance names’ reflecting the subjects and chemicals (or drugs) they mention, respectively. Each substance name mentioned in an article is taken as a name for a chemical trend. In PubMed, MeSH terms are defined in a two-level hierarchy: ‘descriptor names,’ which give the actual MeSH terms, and ‘qualifier names,’ which are used to further detail a descriptor. As MeSH terms, we used only the descriptor names which were annotated as a major topic of an article or had one of their qualifier names annotated as a major topic. When the articles published between 1950 and 2006 are analyzed, these criteria give 23,808 MeSH terms and 174,879 chemical names, which constitute the list of keywords describing the names of our biological and chemical trends.

We assign an impact to each article in PubMed, which is the average of its journal’s IF values between 1999 and 2006 as reported by Thomson ISI. ISI aims to cover more influential journals and has a bias towards high-impact journals. Despite this, there are journals listed in ISI with a 0 impact. Hence, justifiably an IF of 0 is assigned whenever no information is available on a journal in ISI.

In PubMed, each journal has a unique ID (NlmUniqueID), but journal name or ISSN may change in time for several reasons. To map a journal (and its IF) in ISI to articles in PubMed, the journal’s ISSN (as reported by ISI) is searched in PubMed, and all the articles with the NlmUniqueID corresponding to that ISSN are defined to be from the initial journal.

Since MeSH terms have an internal hierarchy defined by NLM, for each MeSH term we give the subfield and field (if available) that are immediately above the term in the tree structure, for a better appreciation of the MeSH term’s classification.

Method

Article trends represent the number of articles published in a field over time. Impact trends represent the average impact of the articles in a field over time. The sum of all article impacts for a specific field is defined as the ‘total impact’ (I) created by the field. We define the total number of articles mentioning a keyword as the keyword’s (field’s) ‘volume’ (V). The ‘impact volume ratio’ (IVR), which is defined by the total impact of a keyword divided by its volume, is a measure of the expected impact generated by each article in a field. As expected, there is a strong positive correlation between I and V values of keywords (MeSH terms, r2=0.83, p<10-6; chemicals, r2=0.90, p<10-6). Using the linear correlation, we compute the ‘expected impact’ (Ie) for each keyword. The value of I/Ie is a normalized measure of a field’s success: a field with an I/Ie value larger than 1 is more successful than randomly expected. Ranks for each of the measures give the percentile of a field compared to other fields.

 
Figure 1. Relation between Impact and Volume for MeSH terms (from 1950 to 2004). Red line represents the linear regression line. The terms above the red line have I/Ie> 1 and, hence, their impact on the scientific community was larger than expected for their Volume.
 

 

 

About

SciTrends.net has been conceived and designed by Murat Cokol and Raul Rodriguez-Esteban at Andrey Rzhetsky's lab, Columbia University. Trends data are available upon request from the authors. Please cite the article (***) when SciTrends is used.

 

Help

For any question please contact us: Murat Cokol ( cokol at hms harvard edu ) or Raul Rodriguez-Esteban ( raul at ee columbia edu ).