Use case cholesterol. TIP Python library in practice

We decided to conduct a study on the use case of cholesterol. Cholesterol is a sterol and an integral part of the cell membrane.
According to field scientist the research topic cholesterol is very diverse. It can even be taken as a paradigm for conversion of food and pharma fields (Bröring et al. 2006):

“For example, the cholesterol-lowering technologies (for example, sterols) are applied for products in both industries [i.e. food and pharmaceutical industries, E.S.]. […] Even though the pharmaceutical industry has traditionally focused on curing, and the food industry on nurturing, the trend towards prevention is witnessed in both sectors, leading to substitute products” (Bröring et al. 2006).

Also, our study confirmed this assumption: The first investigations indicate that research in cholesterol does often occur in the context of cardiovascular disease research, as well as in relation to research in nutrition sciences and studies on digestive system diseases. During the last decades the interest in such fields in co-occurrence to the research in cholesterol differs a lot. Though, since the mid 1970ies a new research topic can be observed coming into the field of cholesterol related studies (Melnychuk et al 2019). Since then research on cholesterol is also often related to gynecological diseases studies.

One hypothesis to explain the emergence of gynecological interest could be the increasing number of women coming into science resulting in the upcoming interest in gynecological research. Applying TIP-Python library we aimed to address the question if the upcoming topic is related to a change of the people who are performing the research. Do these scientist belong to a comparable group of people just shifting in topics due to external reasons or at random? Or does a change in the social group of researchers is related to a change of research questions?

Foundation of data

Our data set covers publications from 1940ies to 2018 annotated with Mesh-Terms. Besides Medline – the world leading database for medicine of US National Library of Medicine (NLM) – also other databases use MeSH vocabulary for annotating content. To include those publications as well (such as collections of Max Rubner Institut and Friedrich-Löffler Institut ), we chose ZB MED Knowledge Environment that includes Medline along other life science databases.
We compiled a list of MeSH terms developed by term frequency inverse document frequency method (tf-idf). To do so we used journals highly regarded in the research field of nutrition science (such as AMERICAN JOURNAL OF CLINICAL NUTRITION, EUROPEAN JOURNAL OF NUTRITION, JOURNAL OF NUTRITION, FOOD \& FUNCTION, JOURNAL OF AGRICULTURAL AND FOOD CHEMISTRY, JOURNAL OF DAIRY SCIENCE) as well as such journals from the pharmaceutical science (such as BRITISH JOURNAL OF PHARMACOLOGY, INTERNATIONAL JOURNAL OF PHARMACEUTICS, ANTIMICROBIAL AGENTS AND CHEMOTHERAPY, JOURNAL OF PHARMACOLOGY AND EXPERIMENTAL THERAPEUTICS). From those typical journals we extracted the annotated Mesh terms. Doing so, we compared the Mesh-Terms in the way to filter for those that are significant for its field, food or pharma, but do not appear in the other section. We calculated an indicator that produces a value between 0 and 1, which expresses the just described capacity. A term highly relevant for one section but with few relevance in the other gains a value close to 0, while a term highly relevant for both section gets a value close to 1.
This list of 2036 MeSH terms limits down food-pharma related papers to a dump of 14.467.574 papers. With a list of 278 MeSH-terms we created a second dump of cholesterol related papers of about 8.287.899 Mio publications. Applying the one and only MeSH-term “Cholesterol”, we created a third and “small cholesterol dump” of 99.333 papers annotated with only “cholesterol” as MeSH-term.

Examination of hypothesis

Applying the Pubmed-Id and DOI we requested the Wikidata-API using SPARQL for authors of the cholesterol related scientific paper in Wikidata. If they are listed, we request the API for the identifier of the specific author(s). In Wikidata this identifier is called Q-Nr/Q-Id.

If the hypothesis – that the increasing participation of women lead to an change in research topics – could be strengthen by quantitative data, this would become a strong argument for a more diverse composition of the scientific staff which could serve the people in its broad variety much better.
Such deliberations in mind we started the enrichment of cholesterol research papers with metadata of authors. With regard to the “big cholesterol dump” of we were able to identify 286.741 matches of authors and papers in Wikidata data base using of Pubmed-Id (december 2019). Based on DOI we determined more than 90.000 matches in Wikidata. As papers often do have multiple authors, just like many authors publish multiple papers, we detected combinations or matches of publications with authors. Thereby, it became possible to shift from the publication level to the level of people acting within the scientific field. Altogether, these matches include 99.409 unique authors and 98.810 unique articles. After all, it is to question if this thin data basis is sufficient to conduct a study.

If you do, you can analyses the data and find out the affiliations where researchers worked who published on cholesterol. Its most common are: University of Copenhagen, Karolinska Institute, King’s College London (see table “registered affiliations of authors in Wikidata).

registered affiliations of authors in Wikidata (big cholesterol dump)
AffiliationNumber of affiliations of authors
University of Copenhagen, Q1862851749
Karolinska Institute, Q2195641613
King’s College London, Q245247 1293
University College London, Q1931961267
University of Oxford, Q344331155
University of Helsinki, Q286951147
University of Leeds, Q503424 1140
University of Cambridge, Q35794 1054
Imperial College London, Q1890221041
University of Bristol, Q459506 982

These top ten institutes employing researchers dealing with cholesterol related topics are quite the same as the top ten of the broader food-pharma dump. Including one exception: The University of Bristol which is specific for the Cholesterol affiliated paper.
Also the analysis of the gender allocation shows a very high similarity in the structure of the cholesterol dump and the food-pharma dump: From the cholesterol data we find in the 1950ies 4 paper written by females, 20 written by men and 0 written by trans-people. In the 1060ies it is: 0 – 1 – 0, 1970ies: 195 – 2168 – 0, 1980ies: 46 – 218 – 0, 1990ies: 6533 – 23294- 0, 2000ies: 15738 – 42652 – 0. The progress is nearly the same as for the food-pharma dump.

Publications by gender and decade (of registered authors of registered papers
decadewomenmen
1950ies420
1960ies 01
1970ies1952168
1980ies46218
1990ies653323294
2000ies1573842652

The “small cholesterol dump” generated by the only MesH term “cholesterol” delivers more interesting data. Of the 99.333 papers 96.207 are registered in Wikidata (91 only by DOI). matches of authors and papers both registered as Wikidata items are at least 2.283. Among then we found 1.183 distinctive authors ad 1.678 distinctive articles.

There are quite different affiliations related to cholesterol-papers compared to the broader perspective of the “big cholesterol dump”. University of Oslo, Stanford University and University of Milan appear under the first five affiliations where most authors are or had been engaged.

Registered affiliations of authors in Wikidata (small cholesterol-dump)
AffiliationNumber of affiliations of authors
Karolinska Institute, Q21956492
University of Helsinki, Q2869543
University of Oslo, Q48615635
Stanford University, Q4150633
University of Milan, Q4621021
University of Copenhagen, Q18628517
University of Gothenburg, Q37152215
Cornell University, Q4911514
King’s College London, Q24524714
University College London, Q19319613

Concerning the question of women joining the field we see a concrete rise of women joining the field in the data of the “small cholesterol dump” as expected since the 70ies. The share of female authors is rising from 2.9\% in the 70ies, to 15.1\% and 19.1\% in the 80ies and 90ies to 21.4\% in the 2000 and reaches 38.8\% in the recent years from 2010-2018.

After all: The data is not sufficient yet for deeper analyses. The hypothesis on gender can neither be rejected nor accepted on this data basis so far.

References

Bröring, S., Cloutier, L. M., & Leker, J. (2006). The front end of innovation in an eraof industry convergence: evidence from nutraceuticals and functional foods.R&DManagement,36(5), 487–498. Retrieved 2019-07-24, fromhttps://onlinelibrary.wiley.com/doi/abs/10.1111/j.1467-9310.2006.00449.xdoi: 10.1111/j.1467-9310.2006.00449.x

Melnychuk, T., Galke, L., Seidlmayer, E., Wustmans, M., Tochtermann, K., Förstner,K. U., . . . Schultz, C. (2019). Analyzing scientific dynamics – does machinelearning help to predict scientific convergence based on bibliographic data?.