DeePathNet: A Transformer-Based Deep Learning Model Integrating Multiomic Data with Cancer Pathways.

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning in oncology

Background:

• Multiomic data analysis is crucial for cancer diagnosis and prognosis.
• Traditional machine learning models often overlook biological network information.
• Integrating domain knowledge, like biological pathways, can improve omic data analysis.

Purpose of the Study:

• To develop an explainable deep learning model, DeePathNet, that integrates cancer-specific pathway information into multiomic data analysis.
• To enhance the prediction of drug response and cancer classification using integrated multiomic and pathway data.
• To enable pathway-level biomarker discovery for improved cancer research.

Main Methods:

• Development of DeePathNet, a transformer-based explainable deep learning model.
• Integration of cancer-specific pathway information with multiomic data.
• Validation using large datasets: ProCan-DepMapSanger, Cancer Cell Line Encyclopedia, and The Cancer Genome Atlas.

Main Results:

• DeePathNet outperforms traditional methods in predicting drug response.
• DeePathNet demonstrates superior performance in classifying cancer type and subtype.
• The model enables effective biomarker discovery at the pathway level.

Conclusions:

• DeePathNet effectively combines biomedical knowledge with deep learning for advanced cancer analysis.
• The model significantly enhances the power of data-driven approaches in cancer research.
• DeePathNet offers a novel tool for improving cancer diagnosis, prognosis, and treatment strategies.