Identifying gene-environment interactions for prognosis using a robust approach

Area of Science:

• Biostatistics
• Genomics
• Cancer Research

Background:

• Prognosis is crucial for complex diseases, influenced by genetic (G) and environmental (E) factors, including gene-environment (G x E) interactions.
• Prognosis data can exhibit contamination or mixture distributions, leading to biased estimations if not addressed.
• Existing methods may struggle with data heterogeneity, impacting the accuracy of prognostic models.

Purpose of the Study:

• To develop a robust statistical model for disease prognosis that accounts for data contamination and mixture distributions.
• To enhance the accuracy of prognostic predictions by incorporating gene-environment interactions.
• To introduce a penalized accelerated failure time (AFT) model with a novel loss function for improved estimation and marker selection.

Main Methods:

• Utilized an accelerated failure time (AFT) model framework.
• Proposed an exponential squared loss function to handle data contamination or mixture distributions.
• Employed a penalization approach for regularized estimation and marker selection, implemented via coordinate descent (CD) and minorization maximization (MM) algorithms.

Main Results:

• The proposed method demonstrates robust performance in the presence of data contamination or mixture distributions, outperforming nonrobust alternatives.
• Achieved comparable or superior performance to existing robust methods like quantile regression in specific scenarios.
• Successfully applied to The Cancer Genome Atlas (TCGA) lung cancer data, identifying significant gene-environment interactions and stable prognostic markers.

Conclusions:

• The developed robust AFT model effectively addresses data contamination and mixture issues in prognostic analysis.
• The method provides reliable identification of gene-environment interactions and prognostic markers, with implications for cancer research.
• This approach offers a valuable tool for improving prognostic accuracy in complex diseases with heterogeneous data.