Identification of usual interstitial pneumonia pattern using RNA-Seq and machine learning: challenges and solutions

Area of Science:

• Pulmonary Medicine
• Genomics
• Bioinformatics

Background:

• Idiopathic pulmonary fibrosis (IPF) diagnosis relies on identifying the usual interstitial pneumonia (UIP) pattern.
• RNA sequencing (RNA-Seq) offers potential for objective diagnostic classification.
• Challenges include limited sample size, biological/technical heterogeneity, and batch effects.

Purpose of the Study:

• Develop and validate an RNA-Seq-based classifier for UIP pattern detection.
• Address limitations of small sample size and sample heterogeneity.
• Improve diagnostic accuracy for IPF.

Main Methods:

• Utilized RNA-Seq data to train a classifier for UIP pattern identification.
• Employed cross-validation and in silico mixed samples for model development and optimization.
• Selected a penalized logistic regression model for its reproducibility.

Main Results:

• The classifier achieved an area under the curve (AUC) of ~0.9 for UIP classification in cross-validation.
• Optimized decision boundary achieved ≥85% specificity.
• The final model demonstrated 70% sensitivity and 88% specificity on the test set.

Conclusions:

• RNA-Seq-based classifier effectively addresses challenges in UIP pattern classification.
• The developed classifier is accurate and robust for diagnosing IPF.
• Methodologies successfully managed sample size limitations, heterogeneity, and batch effects.