Assigning species information to corresponding genes by a sequence labeling framework

Area of Science:

• Bioinformatics
• Computational Biology
• Text Mining

Background:

• Accurate species assignment to genes is crucial for gene normalization in text-mining.
• Current heuristic rule-based methods for gene-species co-occurrence have suboptimal accuracy.

Purpose of the Study:

• To develop a high-performance method for automatic species assignment to genes.
• To improve the accuracy of linking gene mentions to species information in research articles.

Main Methods:

• Developed a novel deep learning-based framework for identifying gene-species relationships.
• Treated species assignment as a sequence labeling task, reducing the number of evaluated gene-species pairs.
• Implemented a deep learning model to identify relationships between genes and species.

Main Results:

• The deep learning approach significantly outperformed the rule-based baseline method.
• Achieved a notable increase in accuracy for species assignment, ranging from 65.8% to 81.3%.
• The novel sequence labeling framework proved more efficient than traditional binary classification.

Conclusions:

• The proposed deep learning framework offers a significant advancement in automatic species assignment for gene normalization.
• This method provides a more accurate and efficient solution for processing biological literature.
• Source code and data are available for species assignment, promoting reproducibility and further research.