From Predicting Cancer Treatment Response to Identifying Novel Therapeutic Targets using Graph Neural Networks

Area of Science:

• Oncology and Computational Biology
• Genomics and Bioinformatics

Background:

• Cancer treatment resistance significantly impacts patient survival, necessitating predictive models.
• Understanding resistance mechanisms is vital for developing effective therapeutic strategies.

Purpose of the Study:

• To introduce TIGENet, an interpretable model for predicting cancer therapy response.
• To identify potential therapeutic targets and key genes involved in treatment resistance.

Main Methods:

• Integration of a variational autoencoder for dimensionality reduction.
• Application of a Graph Neural Network model for predicting patient responses.
• Utilizing a graph explainer for identifying influential genes.

Main Results:

• TIGENet successfully predicts patient response to cancer treatments using transcriptomic and clinical data.
• Identified key molecular factors and genes associated with treatment resistance in breast cancer.
• Highlighted influential genes driving model predictions for enhanced interpretability.

Conclusions:

• TIGENet offers an interpretable approach to predicting cancer therapy response and resistance.
• Findings support the advancement of personalized therapeutic interventions in oncology.
• The model aids in identifying patients at risk of resistance and relevant molecular drivers.