Visit to EuropePMC/EBI 2025-11-19

[petermr]

Topics to discuss at EBI when PMR and Shweata Hegde visit

[rkclimate20]

Detailed Program schedule for semantic Climate demonstrations 10:45 - 12:15 UTC (16:15 – 17:45 IST)

• 10:45 - 10:55 UTC (10 min) Renu Kumari (Introduction of semanticClimate)

• 10:55 - 11:05 UTC (10 min) Peter Murray-Rust (Overview of technology)

• 11:05 - 11:20 UTC (15 min) Priti (Supervised knowledge acquisition)

• 11:20 - 11:35 UTC (15 min) Udita (Unsupervised analysis)

• 11:35 - 11:50 UTC (15 min) Shaik Zainab (Creating knowledge graph for Climate Justice)

• 11:50 - 12:00 UTC (10 min) Nidhi (How to explore Europe PMC : Video presentation)

[petermr]

Agenda

From
Summer Rosonovski
(All UK times are UTC - I have added IST)

Dear Peter,

Draft agenda

• 09:30–10:30 Europe PMC overview 1500-1600 IST

• 10:45–12:15 Semantic Climate demonstrations 1615-1745 IST

• 12:15–13:00 Lunch 1745 - 1830 IST

• 14:00–15:00 /Joint discussion on next steps
  and future collaboration 1930 - 2030 IST

NOTE: Please check IST.

[petermr]

Issues for discussion

Please add more or comment on these
(some may be covered already in EPMC presentation).

From semanticClimate

• Other repositories not covered in morning session:
• Dictionaries and encyclopedias
• Term-based searching including weights
• VectorStore and other "AI" components
• Online communities and Outreach
• Possible online workshop

I will open separate comments for each

[petermr]

Other Repositiories

OpenAlex

semanticClimate has built an interface to OpenAlex and we have ongoing discussions

• how does OpenAlex metadata content differ from EPMC?
• how does OpenAlex fulltext content differ from EPMC?

biorxiv and medrxiv

semanticClimate has built an interface to OpenAlex and we have ongoing discussions. There is a throttling limit that can be bypassed on request.

• do these repos have anything that EPMC does not?
• what is the timescale of posting?

Redalyc, AmeliCA and SciELO

semanticClimate has built an interface to Redalyc and SciELO.

• does EPMC harvest any/all these repos?

what other repos are of interest?

• theses
• INRxiv
• AfricaRxiv

are there restrictions on the use of other repos?

[petermr]

Dictionaries, encyclopedias and annotations

We use dictionaries of terms for:

• supervised search
• annotation
• linking to other resources

An encyclopedia is a superclass of a dictionary enhanced with:

• textual descriptions and images
• cross links in a knowledge graph

encyclopedias can be used wherever dictionaries can. Terms are either taken from public terminologies (e.g. IPCC climate, supervised) or extracted from a corpus. Terms become entries in encyclopedias through:

• formal definitions
• descriptions
• figures, tables, etc.
• indexing into Wikidata (with ID)

Q for EuropePMC:

• where do your search terms come from?
• what's the licensing?

[petermr]

term-based queries and weights

[petermr]

term weights in articles

weights

A drawback of many search systems is that they are binary - term does or does not occur. We would like weights so we can

• find articles with frequent occurrence of the term.
• order retrieved articles in order of term frequency
• negatively weight unwanted terms

Q how much of this is currently possible? How are returned articles ordered for relevance?

multiple terms in queries

We would like to query with large numbers of terms, e.g.:

• plant names
• chemicals
• countries

automatic generation of queries from (un/weighted) list of terms

This could be done with computed Boolean queries but there was a limit to the number of OR terms in EPMC queries.

• Is that still true?
• is there a better way?
• search by document fulltext?

[petermr]

VectorStore and other "AI" components

[petermr]

AI developments

You mentioned when we met that EPMC was developing a vector store. We'd love to know more

[petermr]

Online communities, Outreach, Online workshops

[petermr]

Outreach, etc.

We have experience of running events IRL, hybrid, and fully virtual. Can we explore opportunities here?

[rkclimate20]

Online Communities

• we are the group of people across the globe.
• Online and offline Internship for 3-6 months or longer.
• Working on creating, distributing and using software.
• developing resources for climate knowledge and automated literature review.
  Read more

Outreach

• Interaction with school students and encouraged them to explore science and philosophy of climate change in a fun and interactive way.
• Provide Biodiversity knowledge to the students with board and card games.

Workshops and Hackathons

• semanticClimate organizes online and offline workshops for Ph.D. students, Scientists, researchers, academicians, librarians.
• demonstration of Installation and how to use semanticClimate toolkit
• Extracting keywords/phrases from reports and literatures
• Constructing climate knowledge graphs and encyclopedia.
• Creating corpus of scientific literature
• Automated Literature Review

Participation in International Events

• FSCI 2025: semanticClimate has conducted three days online course on automated literature review and computer vision
• OPEN ACCESS WEEK 2025: demonstrated the use of semantic tools to create corpus and encyclopedia
• DESTiNY Hackaton 2025: Participated in the two days hackathon in PIK Germany

Read more about the events and outreach

semanticClimate Resources

[petermr]

Overview of Technology

Peter Murray-Rust, semanticClimate and University of Cambridge
MACiE at EBI
Primary goal: knowledge to help climate science and its users

By-product: toolkit for knowledge discovery and re-use

NOTE: unlike "AI", all material has provenance, is Open and is "trustable". Code: Apache-2, Documents CC BY or CC0

experts and non-expert

• biomedical scientists (e.g. phytochemistry)
• digital humanities (e.g. climate justice)
  JEP article
• undergraduates ("sustainability" in Cambridge Chemistry)
• late high-school ("Climate Academy")
• U3A (University of Third Age)
• policy and activism

key features

• truthful trustable knowledge is a key component of tackling current global problems.

• automation of search and processing

• community creation and outreach

• tools/materials are tailored for non-experts as well as experts

• learning-by-doing and self-learning

• semantically enhanced terminology (dictionaries and encyclopedias)

• supervised and unsupervised approaches

  • supervised/targeted - you (think you know) what you want
  • unsupervised/exploratory - you have to explore a (new) subject

  semanticClimate supports both. Unsupervised often means creating a larger corpus than necessary and exploring it

Corpora and Encyclopedias ("Dictionaries")

Automatically generate all components within minutes (time depends on downloads, and term-extraction)

[gilienv]

Topics to discuss at EBI when PMR and Shweata Hegde visit

[gilienv]

Thank you Peter, the program looks full! Plesae share Zoom link here, I will try to join if I am back in time from our ANRF meeting today :)
Gita

[petermr]

It would be GREAT to see you.
The current team has created material with little steering and have valuable presentations.

[mrchristian]

Good luck team #semanticClimate

[ShweataNHegde]

2025-11-19 Meeting Notes

Team Members: Madhu (tech), Muhamed (tech), Daljit (user experience)

Peter's Initial Thoughts

• Making climate knowledge accessible through EPMC
• Reaching out beyond academics: digital humanities, general publics, University of the Third Age, schools in India and Srilanka
• We need trustworthy information

EPMC Team

• People usually come from Google Scholar.
• ChatGPT is now number 3 in bringing people into EPMC. Put public services like EPMC under an overload because of bots crawling.

EPMC, Literary Services Team

• Open, full-text, engaging users, enabling contributors, and integrating with other research outputs.
• Life sciences from trusted sources: 1.4 Billion API requests
• Accessing content: Abstract and Full text, Articles API, Grants API, Annotations API, bulk download of content
• Grant Finder
• ORCID claiming tool
• Open Citation Network
• Accepted manuscript: Europe Plus, Open Access Publication, open source code
• Deep data integration, AI innovation?

API

• More info available here: https://europepmc.org/developers
• Articles Restful API: https://europepmc.org/RestfulWebService (with Swagger interface)
• Rate limit: 500 requests per minute
• Annotations API: Gives access to all text-mined annotations within abstracts and full texts

Annotations

• Biological entities: chemicals and molecules, organisms, experimental methods, gene-disease relations, gene/protein, diseases, gene ontology, metagenomics, data resources, data citations.
• billion annotations: abstracts and full text
• Data accession mining: DNA and RNA, proteins, cross-domain, gene expression, cell lines, imaging
• Supporting metagenomics and drug discovery research: Enriching metagenomics data, identifying drug-disease association (machine learning pipeline). Drug-disease associations. Links into open targets.
• Open Targets: Improve the drug discovery process.

Open Source software co-development

• Developing open infrastructure
• Promoting responsible research assessment
• Supporting drug discovery.

How Generative AI Changes Search

• ChatGPT and Perplexity: Search and Chat.

Focus Groups provide insights into user needs

Focus groups: Mostly early-career researchers and librarians. Twitter and Bluesky, Personal Connections.

• People accidently discover Europe PMC; default to PubMed Centre.
• Search assistance: Search has to be better.
• A/B Split test: Save and create alert, full-text link outs, and result sorting.
• User Analytics: Search improvements.
• Expose data and bioentities from supplemental files: 48.5 million annotations.
• Replacement of diction-based annotations. Do machine learning.

Restructured Machine Learning Pipeline Architecture

• Unified Pipeline Design, Centralised Article Classification, Interoperable Components, FAIR-Compliant Outputs.
• Dictionaries are not public yet

Support Open AI models

• Create a Vector Database of Europe PMC for experimentation and feedback from others
• Enhance access to open content for AI tools
• Establish collaborations with experts in the field and engage with the AI community
• Using summarisers (brings down cognitive overload)

Improved API aggregator

• Allows users to traverse rich data in a single query with a single API.
• Introduce provenance in data structure.
• Strengthen outreach to the programmatic user community.

semanticClimate's presentation

semanticClimate: 5-year-old organisation, youth champions

• Workshops on literature reviews, outreach to school students
• Presentation of the workflow today

Text-mining tool

https://gitlab.ebi.ac.uk/literature-services/public-projects/textmining-api

[ShweataNHegde]

• Potential Liasing
• Check back in every year according to the road map.
• Look at the annotation submission.
• Make quick online videos with Summer.

Some general advice

Using singularity for data pipelines.
Search POST