Enabling Scientific Reproducibility through FAIR Data Management: An ontology-driven deep learning approach in the...

Area of Science:

• Neuroscience
• Computational Biology
• Data Science

Background:

• Scientific reproducibility is crucial for advancing research, especially in neuroscience.
• Accessing existing study data for replication is often challenging, requiring manual searches and author contact.
• Neuroimaging and electrophysiology are key data modalities in neuroscience research.

Purpose of the Study:

• To develop an integrated search platform, NeuroBridge, for discovering relevant neuroscience study datasets.
• To enable researchers to test hypotheses and replicate findings more efficiently.
• To overcome the difficulties associated with traditional data discovery methods.

Main Methods:

• Developed a metadata ontology based on W3C PROV specifications for semantic annotation.
• Created a corpus of semantically annotated scientific papers using the Inception tool.
• Employed a Bidirectional Encoder Representations from Transformers (BERT) deep learning model trained on the annotated corpus.
• Utilized the NeuroBridge ontology and BERT model to identify and rank candidate datasets for neuroimaging studies in drug abuse and schizophrenia.

Main Results:

• The NeuroBridge system demonstrated superior performance compared to the PubMed Central (PMC) search engine.
• The combined ontology and deep learning approach showed significant trainability and transparency.
• Automated identification of candidate datasets for neurocognitive assessment was achieved.
• Accuracy and recall were evaluated by independent users.

Conclusions:

• The NeuroBridge platform effectively facilitates the search for neuroscience datasets, enhancing research reproducibility.
• The integration of a specialized ontology with deep learning models offers a transparent and trainable solution for data discovery.
• NeuroBridge represents a significant advancement over traditional search methods for neuroscience research data.