Deliverable 4.9: Evidence-based Knowledge Base Creation: Initial results

This deliverable presents the initial results achieved during the first 12 months of task T4.6 “Evidence-based knowledge graph creation and exploration”, of Work package 4 “Multimodal Analytics and Learning Platform” of the HEREDITARY project.

The work conducted during the first period of research includes a review of state-of-theart approaches and the development of novel unsupervised and supervised models for extracting useful information about gut-brain interplay from biomedical literature, which can provide valuable insights for healthcare. This report presents initial results from the development of methods for information extraction from biomedical literature (named entity recognition, relation extraction) and methods for building Knowledge Bases on top of the extracted entities and relations.

The report presents a description of three main results: PubMiner AI1 and solutions developed by the consortium for the GutBrainIE Task at CLEF 2025 BioASQ Lab, organized by the HEREDITARY project as part of Task 5.6. The main source of the data used for the development of the proposed approaches is PubMed citations. PubMiner AI provides an end-to-end solution based on large language models (LLMs) that covers the full spectrum of functionalities - information extractions (NER and RE) and transformation into Knowledge Graphs (KGs). The solutions developed for GutBrainIE Task at CLEF 2025 BioASQ Lab, provide approaches for NER and RE from PubMed articles covering 13 categories of entities and 25 relation types with focus on the GutBrain axis. The evaluation of the proposed models demonstrates their capabilities to extract gut-brain interplay-related information from PubMed articles. The organised challenge (that will be fully described and reported in Deliverable 5.10 due at Month 24) and the developed tools will bootstrap further research in the area.