Supporting Workflow Reproducibility by Linking Bioinformatics Tools across Papers and Executable Code

Area of Science:

• Computational Biology
• Bioinformatics
• Scientific Workflow Management

Background:

• The increasing volume of biological data necessitates transparent and reproducible computational workflows.
• Linking workflow code to published descriptions is crucial for comprehension, reproducibility, and reuse.
• Current methods lack automated approaches to connect tool mentions in scientific literature with workflow implementations.

Purpose of the Study:

• To present CoPaLink, an automated system for linking bioinformatics tools between scientific text and workflow code.
• To improve the transparency and reproducibility of computational workflows in biological data analysis.

Main Methods:

• CoPaLink integrates three components: Named Entity Recognition (NER) for tool mentions in text and code, and entity resolution using word embedding similarity.
• Utilizes Sentence-BERT for entity recognition and resolution.
• Leverages curated corpora of scientific articles and workflow code with tool annotations.

Main Results:

• Achieved high individual F1-measures (77-90) for NER components.
• Obtained a joint accuracy of 66 for linking tools in Nextflow workflows.
• Demonstrated the effectiveness of CoPaLink in bridging narrative descriptions and code implementations.

Conclusions:

• CoPaLink offers an automated solution for connecting bioinformatics tools across scientific publications and workflow code.
• The approach significantly enhances workflow comprehension, reproducibility, and facilitates tool reuse.
• The developed system and corpora are publicly available to support further research.