This repository contains code, data, prompts and results related to the (semi-)automatic pipeline of Ontology and Knowledge Graph Construction with four different Large Language Models (LLMs): Mixtral 8x22B Instruct v0.1-Instruct-v0.1, GPT-4o, GPT-3.5, Gemini.

This repository contains code, data, prompts and results related to information retrieval from PDFs using multiple LLMs and the RAG approach. Later, results from individual LLM were used to create a hard voting classifier, which enhances the overall quality of the results. LLMs used: Llama-3 70B, Llama-3.1 70B, Mixtral-8x22B-Instruct-v0.1, Mixtral 8x7B, and Gemma 2 9B.

Scripts and pipelines for the large-scale study of Jupyter notebook reproducibility from biomedical publications. Covers mining PubMed Central, locating notebooks on GitHub, executing them in clean environments, and comparing outputs — across 27,000+ notebooks from 2,660 repositories.

A Jupyter Notebook extension that captures and visualizes the provenance of notebook executions over time using the REPRODUCE-ME ontology. Stores execution history as RDF, allows side-by-side comparison of runs, and supports SPARQL querying of experiment histories.

A web-based visualization tool for assessing the reproducibility of Jupyter notebooks stored in GitHub repositories. Displays counts of reproducible, exception-throwing, and result-differing notebooks, with RDF provenance export via ProvBook integration.

A JupyterLab extension that automatically tracks, manages, compares, and visualizes provenance of machine learning notebooks. Identifies relationships between data and models in ML scripts, records datasets and modules used, and enables comparison across experimental runs.

MLProvCodeGen is a JupyterLab extension that automates the generation of machine learning training code while simultaneously capturing fine-grained provenance data according to real-world models. It enables the seamless reproduction of experiments from recorded provenance files and generates relational graphs to provide a clear visual representation of the entire research workflow.

Collaborative Environment for Scientific Analysis with Reproducibility — an end-to-end provenance management framework for scientific experiments. Allows scientists to capture, manage, query, and visualize the complete path of an experiment, covering both computational and wet-lab steps.

Code and gold-standard corpora for Named Entity Recognition (NER) and Relation Extraction (RE) in the biodiversity domain, generated from biodiversity dataset metadata and publication abstracts. Supports training and evaluation of NLP models for ecological text.

A knowledge graph encoding metadata about Jupyter notebook reproducibility at a granular level — notebooks, repositories, cells, outputs, execution environments, and linked publications. Enables semantic querying, provenance exploration, and FAIR sharing of reproducibility evidence.

A generic data model and ontology for representing scientific experiments with full provenance. Extends PROV-O and P-Plan with eight experiment components (Data, Agent, Activity, Plan, Step, Setting, Instrument, Material) to enable end-to-end reproducibility tracking.

An ontology network for the reproducibility of scientific studies, linking foundational and core ontologies covering scientific experiments, machine learning workflows, computational notebooks, and microscopy. Built using a semi-automated approach with ontology matching.

A core ontology for biodiversity that links foundational and domain-specific ontologies, covering taxa, observations, specimens, and ecological relationships. Designed to support semantic interoperability across biodiversity data sources and NLP pipelines.

Dataset from an exploratory survey of researchers across disciplines investigating scientific experiments and research practices relating to reproducibility. Includes questionnaire responses, analysis notebooks, and derived statistical summaries. Published on Zenodo.

A large-scale dataset of 27,000+ Jupyter notebooks from biomedical publications (PubMed Central), capturing reproducibility results, execution environments, error types, and linked metadata. Available as both raw CSV and as a FAIR knowledge graph on Zenodo.

Two gold-standard corpora for Named Entity Recognition and Relation Extraction in the biodiversity domain, generated from biodiversity dataset metadata and publication abstracts. Covers entity types including taxon, location, habitat, trait, and ecological process.