Introduction to Data Management in Exact Sciences

Doctoral College of Brittany

Damien Belvèze

damien.belveze@univ-rennes.fr

University of Rennes

2025-11-12

Advice on data management
Training (data, reproducibility, identifiers)
Support for data management plans
Curation of the Univ-Rennes collection on Recherche Data Gouv
help for making your research source code reproducible

1. Research Data, What Are We Talking About?

a typical experiment : counting the fish in a river

Figure 2

Which Files Are Important to Make Available?

raw_data_fish_counter.csv
intermediate_data.xls
filter1.py
first_draft_submission.pdf
fish_counter_calibration.md
kick_off_report.docx
filter2.py
notebook_experiment.ipynb

final_data_fish_counter.xls
project_presentation_funders.pptx
final_data.csv
study_draft.qmd
january_meeting_partners.docx
fish_counter_instructions_for_use.pdf
gantt_calendar.xlsx

Answers

raw_data_fish_counter.csv
intermediate_data.xls
filter1.py
first_draft_submission.pdf
fish_counter_calibration.md
kick_off_report.docx
filter2.py
notebook_experiment.ipynb

final_data_fish_counter.xls
project_presentation_funders.pptx
final_data.csv
study_draft.qmd
january_meeting_partners.docx
fish_counter_instructions_for_use.pdf
gantt_calendar.xlsx

2. Towards Cumulative, Reliable, and Reproducible Science?

📓 Berners-Lee (2015)

From buried-in-a-PDF data…

Even Optical AI tools are bad at extracting data from PDFs 📓 Edwards (2025)
Mistral failed at converting tables in PDF into markdown (figures truncated)
PDF should be one of the layer not the only one to be communicated
share native format and use reproducible tools such as R

…through Excel (but why Excel would be necessary anyway)

better to share flat files (not workbooks)
have you considered collecting your data in CSL rather than Excel ?

Unlike a script written in a programming language such as Python or R, which documents every step of the process and can be saved, versioned and rerun, an analysis that happens inside a spreadsheet using pointing and clicking is hard to follow and even harder to replicate.

📓 Melchor (2025)

Excel (as every non FOSS tool) itself is prone to errors

Figure 4: CSV vs XLS

📓 Ziemann et al. (2023)

Preserve the raw data

The raw and final data are important to preserve and later deposit in a repository
if your data management is well documented, you don’t need to keep intermediary versions
keep the file that contains your raw data untouched (lock it down in a separate folder), use a duplicate as your working version

…to Linked Open Data

“Lowest temperature in Galway on the fourteenth of January 2020”

concept	ontology	expression
lowest temperature	http://inamidst.com/sw/ont/meteo	Property :temperature
temperature type	Wikipedia	https://fr.wikipedia.org/wiki/Degr%C3%A9_Celsius
Galway	DBpedia	https://dbpedia.org/page/Galway
14th of January	Unix epoch	1578984191

Permanence of Data Access

📓 Gibney & Van Noorden (2013)

3. A Challenge for Open Science

FAIR Principles

Figure 5: FAIR principles

openness / closure

“as open as possible, as closed as necessary”
Default openness
Closure to justify:
- personal data
- intellectual property
- industrial secret
- defense

personal data

The challenge remains to provide access to the dataset to the fullest extent possible. To do this, processes will be used to pseudonymize or anonymize the dataset. Pseudonymization involves removing directly identifying data, or replacing them with pseudonyms. This method carries risks of re-identification through cross-referencing the dataset with other datasets concerning the same population. Anonymization of data is a more subtle method involving, after removing directly identifying data, making the data less precise or adding noise to hinder potential re-identification attempts. It is a compromise between the interests of the individuals whose data are collected and the precision of this data. Tools exist to anonymize data: Arx, Amnesia. Anonymizing data incurs processing costs (human time, performance) that should be anticipated in advance in the budget request.

intellectual property

In France, the regime of data law means that data by default belongs to employers, not producers (researchers). But there may be limitations on sharing this data: - contaminating licenses of certain proprietary materials used to produce data (e.g., medical imaging) - international research projects: which national law will be applied to this data?

Making Your Data Findable

Quality of a directory:

reputation
sustainability (institutional support)
open license
persistent identifier
richness of metadata
curation

discipline	repository
images (SHS)	MediHal
code	Software Heritage via HAL
Bioinformatics	GenOuest
Humanities	Nakala
Mathematics	no repository, see with the RNBM group
environment, hydrology	Data Indores
Earth Sciences	data terra
Marine Sciences data	ifremer, seanoe
medical sciences	INSERM repository on RDG
Ecology, Environment, and Society Data	InDoRES and Cat.InDoRES

Recherche Data Gouv

richness of metadata
curation
national reference (supported by the Ministry)
persistent identifier
significant volume
free of charge
simplified generation of datapapers
RDG sandbox

Are Data Accessible?

In 93% of cases no response or negative response without justification 📓 Gabelica et al. (2022)

Are Data Interoperable?

Which identifiers to use for copper telluride?
registry	identifier
CAS number	12019-52-2
PubChem CID number	6914517
PubChem SID number	24879035
openSMILES identifier	CuCu.CuCu.TeTe
InChI identifier	InChI=1/2Cu.Te
MDL number	MFCD00049727

Documenting the Data

Documentation is the glue that binds a data science project together 📓 Ziemann et al. (2023)
Carefully describe the data and the context of its acquisition (production, collection)
literate programming
describe the data using ontologies

Documenting to Avoid Context Errors

Be precise in describing the context of data production

Figure 7: the importance of data context

Ontologies

discipline	thesaurus
biodiversity	INRAE
environment	GEMET
Biology, Health	MeSH
Mental Health	ascodopsy

directory of thesauri

Reusable Data?

Creative Commons (CC:by)
a license written by a law firm expert in intellectual property that provides for a variety of authorized or prohibited use cases
ODBL
Etalab
no license, do whatever you want with my dataset
CC0
CC:by for everyone except for fossil industries, arms sellers, and Google (📓 Thomas (2023)) text available here.

Ask the students which data mentioned in this slide are appropriate for a dataset.

Answers:

Etalab and ODBL licenses are recommended in France to allow data reuse. Creative Commons licenses are more suitable for research outputs (publications, preprints, posters…) CC0 license is not allowed in France where copyright law requires at least citing the author of the dataset, which is not provided for in this license.

When no license is indicated, one cannot safely reuse the dataset. When there is no license, by default, copyright law applies.

In the context of Open Science, licenses are standardized and the uses they allow are easily stated and understandable. If instead of a license, there is a text that resembles the T&Cs of a web service, it means that the dataset is likely to be difficult to reuse (probably need to contact the producers)

Open Science can be seen by some researchers as a loss of control over their publications and data, and restrictive licenses that hinder reuse for purposes deemed bad by researchers are still in the project stage. These licenses are not usual for data.

Data Management Plan

The DMP summarizes all the choices made for data management
Submit an initial version of a DMP 6 months after signing a contract (ANR, European projects)
DMP OPIDOR
DMP Online

4. Let’s Get Practical

Figures

figure	credits
Figure 2	yanmar https://www.yanmar.com/global/news/2021/04/20/90815.html
Figure 1	Harvard Biomedical Data managemnet
?@fig-data_steps	Tim Berners-Lee
Figure 3	Jon Ippolito
?@fig-data_loss	Gibney, Van Noorden
Figure 5	Willkinson, Dumontier et al.
Figure 6	Sergio Uribe
Figure 4	meme dont l’origine se perd dans la nuit des temps
Figure 7	Ralph Aboujaoude Diaz

Software Used for the Presentation

The presentation was created with free and Open Source software. Thank you to all people who make them alive ❤️ ❤️

Slides

[1] "Quarto version: 1.6.40"

R version 4.5.2 (2025-10-31)
Platform: x86_64-pc-linux-gnu
Running under: Ubuntu 24.04.3 LTS

Matrix products: default
BLAS:   /usr/lib/x86_64-linux-gnu/blas/libblas.so.3.12.0 
LAPACK: /usr/lib/x86_64-linux-gnu/lapack/liblapack.so.3.12.0  LAPACK version 3.12.0

locale:
 [1] LC_CTYPE=fr_FR.UTF-8       LC_NUMERIC=C              
 [3] LC_TIME=fr_FR.UTF-8        LC_COLLATE=fr_FR.UTF-8    
 [5] LC_MONETARY=fr_FR.UTF-8    LC_MESSAGES=fr_FR.UTF-8   
 [7] LC_PAPER=fr_FR.UTF-8       LC_NAME=C                 
 [9] LC_ADDRESS=C               LC_TELEPHONE=C            
[11] LC_MEASUREMENT=fr_FR.UTF-8 LC_IDENTIFICATION=C       

time zone: Europe/Paris
tzcode source: system (glibc)

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

loaded via a namespace (and not attached):
 [1] digest_0.6.37     later_1.3.2       fastmap_1.2.0     xfun_0.53        
 [5] knitr_1.50        htmltools_0.5.8.1 rmarkdown_2.29    lifecycle_1.0.4  
 [9] ps_1.7.6          cli_3.6.5         processx_3.8.3    compiler_4.5.2   
[13] rstudioapi_0.15.0 tools_4.5.2       quarto_1.5.1      evaluate_1.0.4   
[17] Rcpp_1.1.0        yaml_2.3.10       rlang_1.1.6       jsonlite_2.0.0


 quiz : H5P 
 
 repository : Framagit owned and managed by the Fnch pro-FOSS association Framasoft

References

Berners-Lee, T. (2015). 5-star open data.

Edwards, B. (2025). Why extracting data from PDFs is still a nightmare for data experts. In Ars Technica. https://arstechnica.com/ai/2025/03/why-extracting-data-from-pdfs-is-still-a-nightmare-for-data-experts/

Gabelica, M., Bojčić, R., & Puljak, L. (2022). Many researchers were not compliant with their published data sharing statement: Mixed-methods study. Journal of Clinical Epidemiology, 0(0). https://doi.org/10.1016/j.jclinepi.2022.05.019

Gibney, E., & Van Noorden, R. (2013). Scientists losing data at a rapid rate. Nature. https://doi.org/10.1038/nature.2013.14416

Melchor, S. (2025). Six questions to ask before jumping into a spreadsheet. Nature, 644(8076), 569–570. https://doi.org/10.1038/d41586-025-02511-z

Thomas, M., Éric Tannier. (2023, May 17). Se réapproprier la production de connaissance - AOC media. AOC media - Analyse Opinion Critique. https://aoc.media/opinion/2023/05/17/se-reapproprier-la-production-de-connaissance/

Ziemann, M., Poulain, P., & Bora, A. (2023). The five pillars of computational reproducibility: Bioinformatics and beyond. Briefings in Bioinformatics, 24(6), bbad375. https://doi.org/10.1093/bib/bbad375