Ordinal Sparse Neural Networks for Modeling Gene- and Imaging-Environment Interactions.

Area of Science:

• Biostatistics
• Bioinformatics
• Computational Biology

Background:

• Gene-environment (G-E) and imaging-environment (I-E) interactions are crucial for understanding disease etiology.
• Existing methods lack flexibility in modeling ordinal responses like tumor pathological stage.
• There is a need for advanced statistical approaches to analyze complex interactions in biomedical data.

Purpose of the Study:

• To develop a novel neural network-based method for modeling ordinal responses with interaction analysis.
• To enable flexible prediction and variable selection in the presence of G-E and I-E interactions.
• To apply the method to real-world cancer datasets for tumor stage prediction.

Main Methods:

• A neural network architecture with a novel output function for ordinal category prediction.
• Integration of a sparse layer for effective variable selection.
• Utilizing the local quadratic approximation (LQA) algorithm for penalized estimation.

Main Results:

• The proposed method demonstrates competitive performance in both prediction accuracy and variable selection.
• Simulation studies validate the effectiveness of the neural network approach.
• Application to breast cancer (BRCA) and skin cutaneous melanoma (SKCM) datasets successfully identified relevant interactions.

Conclusions:

• The developed method provides a flexible and powerful tool for analyzing G-E and I-E interactions in ordinal response modeling.
• It offers valuable insights into disease mechanisms by identifying significant main effects and interactions.
• The approach has potential applications in personalized medicine and biomarker discovery.