Revealing the Impact of Genomic Alterations on Cancer Cell Signaling with an Interpretable Deep Learning Model

Area of Science:

• Computational biology
• Genomics
• Cancer research

Background:

• Cancer arises from abnormal cellular signaling driven by somatic genomic alterations (SGAs).
• Tumor heterogeneity leads to unique signaling system disruptions.
• Understanding these complex signaling networks is crucial for cancer research.

Purpose of the Study:

• To develop a computational model that interprets cancer signaling systems as causal graphical models.
• To leverage deep learning to predict gene expression changes resulting from SGAs.
• To identify shared functional impacts of SGAs on common signaling pathways.

Main Methods:

• Developed a novel deep learning model named redundant-input neural network (RINN).
• Utilized SGAs as direct inputs to the RINN model.
• Employed causal graphical models to represent cancer signaling pathways.
• Evaluated model performance using area under ROC curves for gene expression prediction.

Main Results:

• The RINN model accurately predicts gene expression based on SGA inputs.
• The model demonstrates the ability to encode the impact of SGAs on cellular signaling.
• RINN successfully identifies shared functional impacts of SGAs affecting common pathways like PI3K, Nrf2, and TGF.
• The model can rediscover known relationships within cellular signaling systems.

Conclusions:

• The RINN model provides a transparent and effective tool for analyzing cancer signaling networks.
• This approach enhances our understanding of how genomic alterations drive cancer progression.
• The RINN model has the potential to uncover novel therapeutic targets by elucidating signaling pathway dysregulation.