Interpretation knowledge extraction for genetic testing via question-answer model.

Area of Science:

• Bioinformatics
• Computational Biology
• Medical Informatics

Background:

• Metagenomic next-generation sequencing (mNGS) is crucial for detecting pathogenic microorganisms in biomedical research.
• Clinical interpretation of mNGS results depends on knowledge bases often built via manual, time-intensive, and potentially biased extraction.
• There is a need for automated methods to efficiently and accurately extract knowledge for mNGS interpretation.

Purpose of the Study:

• To develop and validate an automated approach for extracting pathogenic microorganism knowledge.
• To overcome the limitations of manual knowledge extraction, including time, cost, and subjectivity.
• To enhance the interpretation of mNGS results for clinical applications.

Main Methods:

• Development of a novel question-answer (QA) model approach for knowledge extraction.
• Creation of the MicrobeDB dataset, comprising 3,161 samples from 618 papers on 224 microorganisms.
• Fine-tuning a baseline model with MicrobeDB, enhanced by ChatGPT for data diversity and expansion for increased volume.

Main Results:

• The QA model achieved high performance with an Exact Match (EM) of 88.39% and F1 score of 93.18% on the MicrobeDB test set.
• Ablation studies confirmed the effectiveness of the proposed data augmentation techniques.
• Comparative experiments demonstrated the superiority of the developed method over existing tools like ChatPDF.

Conclusions:

• The automated knowledge extraction method is effective and valuable for interpreting mNGS data.
• This approach offers a significant improvement over manual knowledge extraction processes.
• The developed method has the potential to streamline clinical diagnosis and treatment based on mNGS results.