Species Classification for Neuroscience Literature Based on Span of Interest Using Sequence-to-Sequence Learning...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Computational Biology

Background:

• Neuroscience research generates vast amounts of literature, necessitating efficient methods for knowledge extraction.
• Linking neuroscience findings across species is crucial for advancing brain research, AI, and neurorobotics.
• Current methods struggle to automatically identify species and their research focus within scientific abstracts.

Purpose of the Study:

• To develop an automated system (SpecExplorer) for mining species information from neuroscience literature.
• To distinguish between species mentioned as subjects versus those in a secondary context.
• To facilitate knowledge discovery and resource linking across different research communities.

Main Methods:

• Proposed a sequence-to-sequence classification framework for multi-label species assignment.
• Introduced Hierarchical Attentive Decoding (HAD) to model document structure and extract relevant information (span of interest).
• Created and utilized three datasets from PubMed and PMC corpora with mention-based and semantic-based annotations.

Main Results:

• The SpecExplorer system demonstrated improved performance in species classification tasks.
• Successfully distinguished between primary and secondary species mentions in the literature.
• Enabled novel species-based analyses of brain diseases, cognitive functions, and hippocampal proteins.

Conclusions:

• Automated species identification in neuroscience literature is feasible and valuable.
• The SpecExplorer project offers a scalable solution for knowledge management in neuroscience.
• Findings provide new avenues for species-specific research directions in brain science and related fields.