Variable selection for disease progression models: methods for oncogenetic trees and application to cancer and HIV

Area of Science:

• Biostatistics
• Computational Biology
• Genomics

Background:

• Disease progression models are essential for understanding disease development, incorporating statistical frameworks to manage biological and sampling variations.
• Oncogenetic trees are a popular, flexible model class used for analyzing genetic aberrations in diseases like cancer and HIV.
• Challenges exist in variable selection for oncogenetic trees due to a large number of potential aberrations exceeding model estimation capabilities.

Purpose of the Study:

• To address the gap in variable selection methods for oncogenetic trees.
• To propose and evaluate novel variable selection techniques tailored for oncogenetic tree models.

Main Methods:

• Developed ten variable selection methods for oncogenetic trees, including novel approaches.
• Conducted an extensive simulation study to compare the performance of these methods.
• Applied and validated the methods on real-world cancer and HIV datasets.

Main Results:

• The clique identification algorithm demonstrated superior performance in preselecting relevant events.
• This method effectively identifies events belonging to the largest or maximum weight connected subgraphs.
• Performance was evaluated through simulations and real data analysis.

Conclusions:

• The clique identification method for variable selection is highly effective for oncogenetic trees.
• This approach successfully identifies both frequent and pathway-related important events.
• The findings enhance the reliability and interpretability of disease progression models.