PTBD: a machine learning-based non-invasive diagnostic model for pulmonary tuberculosis using large-scale blood...

Area of Science:

• Biotechnology
• Genomics
• Medical Diagnostics

Background:

• Accurate pulmonary tuberculosis (PTB) diagnosis is crucial for timely intervention and preventing disease spread.
• Current PTB diagnostic methods face limitations in sensitivity, specificity, and practicality.

Purpose of the Study:

• To develop and validate a non-invasive, blood transcriptome-based diagnostic model for PTB.
• To identify novel and robust transcriptomic biomarkers for PTB detection.

Main Methods:

• Utilized a dataset of 2,792 peripheral blood transcriptome samples.
• Integrated top-scoring pair analysis with machine learning for the diagnostic model (PTBD).
• Validated the model on test and independent external datasets.

Main Results:

• The PTBD model accurately distinguished PTB from various other conditions, including healthy individuals, latent TB, pneumonia, lung cancer, and pulmonary nodules.
• Achieved an AUC of 0.869 in the test set and 0.909 in the external validation set.
• Demonstrated consistent performance across diverse populations and clinical conditions, meeting WHO requirements.

Conclusions:

• The developed PTBD tool offers a broadly applicable method for early PTB diagnosis.
• Identified novel transcriptomic biomarkers, including five diagnostic gene-pair expression patterns, representing PTB molecular hallmarks.
• This approach has the potential to reduce the global PTB burden through timely intervention.