A probabilistic pathway score (PROPS) for classification with applications to inflammatory bowel disease

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning in medicine

Background:

• Gene-based machine learning models struggle with noisy data, external validation, and complex diseases.
• Heterogeneous diseases present challenges for classifiers unable to capture varying gene combinations leading to the same phenotype.
• Pathway-based classification offers a robust alternative by using aggregate biological mechanisms.

Purpose of the Study:

• To develop a novel pathway-based classification approach, PRObabilistic Pathway Score (PROPS).
• To improve classification performance for complex diseases by incorporating gene interactions.
• To differentiate between ulcerative colitis (UC) and Crohn's disease (CD), two major types of inflammatory bowel disease (IBD).

Main Methods:

• Developed PRObabilistic Pathway Score (PROPS), a novel pathway-based method using genes to calculate individualized pathway scores.
• Incorporated gene interactions using probabilistic graphical models for more accurate biological representation.
• Applied PROPS to five IBD datasets to distinguish CD from UC, comparing it against gene-based and alternative pathway-based classifiers.

Main Results:

• PROPS demonstrated superior classification performance in distinguishing CD from UC compared to existing methods.
• The method provided biological insights into key pathways differentiating CD and UC.
• The approach effectively handles complex, heterogeneous disease data.

Conclusions:

• PRObabilistic Pathway Score (PROPS) offers a robust and accurate pathway-based classification method for complex diseases.
• Incorporating gene interactions enhances biological representation and classification performance.
• This method shows promise for differentiating similar complex diseases and understanding their underlying mechanisms.