Identifying Crohn's disease signal from variome analysis

Area of Science:

• Genetics
• Machine Learning
• Computational Biology

Background:

• The exact cause of Crohn's disease (CD) remains unknown despite extensive research.
• Genome-wide association studies have identified numerous CD loci, but they offer limited diagnostic value due to small effect sizes.
• Accurate diagnosis of Crohn's disease is crucial for effective management and prevention of disease onset.

Purpose of the Study:

• To develop and validate a machine learning method for predicting Crohn's disease status using exonic variants.
• To identify novel genes and pathways associated with Crohn's disease pathogenesis.
• To improve the diagnostic accuracy and speed for Crohn's disease.

Main Methods:

• Developed AVA,Dx (Analysis of Variation for Association with Disease), a machine learning approach utilizing exonic variants from sequencing data.
• Trained predictive models on a panel of 111 individuals, incorporating person-specific coding variation and accounting for batch effects.
• Applied the trained models to predict CD status in thousands of individuals from independent cohorts.

Main Results:

• AVA,Dx successfully identified known Crohn's disease genes (e.g., NOD2) and highlighted potential new disease-associated genes.
• The method achieved high precision in predicting CD status, identifying 16% of patients with 99% precision (strict cutoff) and 58% of patients with 82% precision (default cutoff) across over 3000 individuals.
• Demonstrated accurate prediction of CD status in individuals from separately sequenced panels.

Conclusions:

• AVA,Dx serves as an effective tool for uncovering Crohn's disease pathogenesis pathways.
• The method functions as a valuable CD risk analysis tool, with the potential to enhance clinical diagnostic time and accuracy.
• Future improvements may involve larger training datasets and integration of additional features like environmental factors and microbiota data.