GE-IA-NAM: gene-environment interaction analysis via imaging-assisted neural additive model

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Gene-environment (G-E) interaction analysis is vital for understanding cancer etiology.
• Traditional regression-based G-E methods may lack flexibility for complex patterns.
• Deep learning models for G-E analysis can be limited by small sample sizes and high dimensionality.

Purpose of the Study:

• To develop a novel method for gene-environment interaction analysis that incorporates pathological imaging data.
• To improve the flexibility and interpretability of G-E interaction models.
• To leverage pathological images to overcome information deficiencies in existing G-E models.

Main Methods:

• Proposed the pathological imaging-assisted neural additive model (GE-IA-NAM).
• Utilized a flexible and interpretable additive network architecture for individualized effects.
• Implemented an assisted learning strategy integrating pathological image information via joint analysis.

Main Results:

• The GE-IA-NAM demonstrated competitive performance in simulations.
• Analysis of lung and skin cancer datasets from The Cancer Genome Atlas confirmed the method's efficacy.
• The model effectively accounts for individualized genetic, environmental, and interaction effects.

Conclusions:

• Pathological imaging can significantly enhance gene-environment interaction analysis in cancer research.
• The proposed GE-IA-NAM offers a flexible and interpretable approach to model complex G-E interactions.
• This method holds promise for improving our understanding of cancer development and progression.