Automatic consistency assurance for literature-based gene ontology annotation

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Literature-based gene ontology (GO) annotation is crucial for understanding gene functions but manual consistency checks are time-consuming.
• Maintaining GO annotation consistency with evolving literature and vocabulary is a significant challenge in bioinformatics.
• Automated methods are needed to ensure the reliability of GO annotations.

Purpose of the Study:

• To formalize biological database annotation inconsistencies and identify distinct types.
• To develop an efficient text mining method for automatically distinguishing consistent from inconsistent GO annotations.
• To evaluate the performance and robustness of the proposed method.

Main Methods:

• Formalization of four distinct types of biological database annotation inconsistencies.
• Application of state-of-the-art text mining models for automated inconsistency detection.
• Evaluation using a synthetic dataset (BC4GO) with manipulated annotations and detailed error analysis.

Main Results:

• The proposed method accurately distinguishes between consistent and inconsistent GO annotations.
• High precision was achieved on confident predictions, validated through error analysis.
• Two models demonstrated robustness against updates in the GO vocabulary.

Conclusions:

• The developed text mining approach effectively identifies GO annotation inconsistencies.
• The method shows significant value in supporting human-in-the-loop curation processes.
• This approach enhances the reliability and maintainability of biological databases.