Assessing Artificial Intelligence (AI) Implementation for Assisting Gene Linking (at the National Library of...

Area of Science:

• Biomedical Informatics
• Computational Biology
• Life Sciences

Background:

• The National Library of Medicine (NLM) manually links millions of articles to gene records, a process crucial for life sciences research.
• Scaling manual indexing to the growing volume of published literature presents a significant challenge.
• Automatic information extraction methods show promise but face integration difficulties in established workflows.

Purpose of the Study:

• To demonstrate the integration of the GNorm2 machine learning model into the NLM's daily workflow for gene identification.
• To evaluate GNorm2's performance in a real-world biocurator environment.
• To assess the impact of GNorm2 on gene identification accuracy, indexing consistency, and efficiency.

Main Methods:

• Integration of the GNorm2 model into the NLM's existing curator workflow.
• Evaluation of GNorm2 by 8 biomedical curator experts using parameters such as gene identification accuracy and interannotator agreement.
• Iterative improvement of the GNorm2 algorithm based on expert feedback to support 135 gene species.

Main Results:

• GNorm2 demonstrated state-of-the-art performance in identifying genes and their species, handling textual ambiguities effectively.
• Key interface modifications were identified to maximize GNorm2's benefit for curators.
• The GNorm2 algorithm was enhanced to include viral and bacterial genes, expanding its coverage.

Conclusions:

• GNorm2 shows significant potential to improve the accuracy, consistency, and efficiency of gene indexing at the NLM.
• The model is currently undergoing full integration into the regular curator workflow.
• Successful integration of GNorm2 is expected to enhance the value of NLM resources for life sciences research.