Cultivating Knowledge Flow in Open Chemistry

Making knowledge visible: the Wikipedia P703 module

Knowledge needs channels, not just reservoirs.

One important step was enabling Wikipedia articles and infoboxes to directly access found in taxon (P703) relationships from Wikidata in an efficient way.

This sounds like a small technical detail. It is not.

The idea had circulated quietly for years, in hallway conversations, chats, and conferences. I also mentioned it during the WikiCite 2025 conference last August, but it took time before conditions were right. Infrastructure work rarely happens on schedule. It almost never happens on stage. It happens in version histories, talk pages, and tiny edits that fix one Lua bug, enabling thousands of articles to improve forever.

Once data is in Wikidata, it can flow into thousands of chemical articles, in dozens of languages, without duplication or translation overhead. It becomes visible to non-experts, students, and readers who will never see a database interface. This is how open data becomes public knowledge.

These modules are now available on multiple Wikipedias:

• English : Module:P703
• French : Module:P703
• German : Modul:P703
• Italian : Modulo:P703

If you speak another language, feel free to copy them and increase their use.

Initially, I tried to reuse existing modules, and quickly learned why module reuse across Wikipedias is famously difficult. Each wiki evolves its own ecosystem of dependencies, conventions, and technical debt. So the modules were written fully contained, independent of language-specific infrastructure. Only lines that need to be changed are the language-specific translations at the top of the module, trying to follow 18n.

These modules are not a new website or a new interface. They are simply better plumbing.

By default, only 5 organisms are shown, keeping the text clean and readable. But the magic is in how the module handles the rest: each taxon links to its Wikipedia article, and if no article exists in the current language, the module gracefully redirects to an equivalent page in another wiki. A beautiful example of this is war for Quassia africana.

Reaching non-experts

Not everyone wants to learn SPARQL.

I have heard this sentiment countless times. Regardless of personal preferences, if we want the data to truly live, it must reach as many people as possible.

Small tools like the LOTUS Wikidata Explorer help lower that barrier. It is imperfect. It is still growing. But it already allows chemists, curators, and students to access and export data in formats they can actually use.

The principle is simple: knowledge only flows when it reaches people. The data must be seen, explored, and reused. Only then does it fulfill its purpose.

On a more technical note, the LOTUS Wikidata Explorer leverages the powerful IDSM endpoint (Galgonek and Vondrášek 2021), which allows for chemical similarity searches thanks to Sachem (Kratochvíl et al. 2018). Its speed for large-scale queries could never have been reached without QLever (Bast and Buchhold 2017), and chemical depictions come from CDK Depict.

It returns structured metadata for traceability and reproducibility, together with hashes that uniquely identify the query and its results. It can be queried programmatically via simple API calls, for example ?taxon=Gentianaceae or ?smiles=c1ccccc1&formula_filter=true&f_state=required. It works directly in the browser without requiring heavy dependencies, almost everything works out of the box.

Alternatively, users can take advantage of a local version, for example to extract a small, personal knowledge graph in TTL format, or to export all or selected LOTUS data in more chemistry-friendly formats such as JSON or CSV, ready for analysis, visualization, or integration into other workflows.

For example, to export a complete or filtered snapshot locally:

uvx \
  --from \
  git+https://github.com/adafede/marimo \
  lotus_wikidata_explorer \
  export \
  --taxon "*" \   # to get all taxa, else "Gentianaceae", for example
  --format csv \  # also supports json, ttl
  --output 20260119_lotus.csv.gz \
  --compress \
  --verbose

And if you are curious where halogenated compounds appear most often, you can simply ask:

xan dedup 20260119_lotus.csv.gz \
  --select compound_inchikey,molecular_formula,taxon_name | \
  xan select compound_inchikey,molecular_formula,taxon_name | \
  xan filter "contains(molecular_formula, 'Br') or
              contains(molecular_formula, 'Cl') or
              contains(molecular_formula, 'F') or
              contains(molecular_formula, 'I')" | \
  xan filter '!contains(molecular_formula, "Fe")' | \
  xan freq -s taxon_name | \
  xan hist -l value -R

Histogram for taxon_name (bars: 11, sum: 11,604, max: 10,060):

Streptomyces            |   374   3.22%|■■■■                                                                                                   |
Laurencia dendroidea    |   314   2.71%|■■■■                                                                                                   |
Laurencia obtusa        |   184   1.59%|■■                                                                                                     |
Pseudoceratina purpurea |   112   0.97%|■■                                                                                                     |
Aplysia dactylomela     |    99   0.85%|■■                                                                                                     |
Nostoc                  |    99   0.85%|■■                                                                                                     |
Laurencia nipponica     |    95   0.82%|■                                                                                                      |
Portieria hornemannii   |    94   0.81%|■                                                                                                      |
Lyngbya majuscula       |    88   0.76%|■                                                                                                      |
Chaetomium globosum     |    85   0.73%|■                                                                                                      |
<rest>                  |10,060  86.69%|■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■|

Blue Obelisk

• Following some ideas he had to improve Scholia Chemistry, Egon initiated the Blue Obelisk Wikidata Chemistry Curation project: https://blueobelisk.github.io/wikidata-chemistry-curation/. I then joined and contributed to some parts, maybe you will find out which ones!
• Contributions were also made to the Blue Obelisk IUPAC Names project (also led by Egon): https://github.com/BlueObelisk/iupac-names, integrating Wikidata-derived name-compound pairs. See https://github.com/Adafede/wd-labels-to-iupac and https://chem-bla-ics.linkedchemistry.info/2025/08/09/one-million-iupac-names-4.html (Willighagen 2025) These bridges allow names and identifiers to circulate consistently across systems.

Reactions flow

Recently, I also contributed to improving how chemical reactions are modeled in Wikidata.

Previously, many reactions were modeled as instances of "chemical reaction", which violated disjointness, see (Doǧan and Patel-Schneider 2025).

The introduction of type of chemical reaction now allows reactions to be classified more precisely, while preserving their hierarchy using subclass of.

Wikifunctions

Out of curiosity, I also made a small contribution to chemistry-related functions in Wikifunctions: Z30950. It validates CAS Registry Numbers. It does one thing, and it does it reliably.

It is tiny. But it is a seed.

References