Identifying important gene signatures of BMI using network structure-aided nonparametric quantile regression

Area of Science:

• Genomics
• Biostatistics
• Bioinformatics

Background:

• Identifying genetic risk factors for complex traits like higher body mass index (BMI) is crucial.
• Existing methods for incorporating prior biological information often rely on restrictive assumptions.
• There is a need for robust statistical frameworks that leverage biological networks for gene association studies.

Purpose of the Study:

• To develop a novel nonparametric additive quantile regression method with network regularization.
• To effectively incorporate prior biological network information into the identification of genomics risk factors for BMI.
• To address limitations of existing methods by relaxing assumptions on the mean function.

Main Methods:

• Nonparametric additive quantile regression framework.
• Network regularization using the total variation norm on linked genes.
• B-spline basis expansion for modeling nonlinear associations.
• Group Lasso penalty for achieving model sparsity.
• Efficient computational procedures for model optimization.

Main Results:

• Simulation studies demonstrate superior performance in identifying true positive genes and reducing false positives compared to alternative methods.
• Application to the Framingham Heart Study microarray gene-expression dataset identified significant gene associations at the 75th percentile of BMI.
• The method successfully leverages known biological network information for enhanced gene discovery.

Conclusions:

• The proposed method offers an efficient and flexible approach for identifying outcome-associated variables using network information.
• It advances nonparametric additive quantile regression by integrating biological network structures.
• This framework is valuable for gene-environment interaction studies and precision medicine.