PathX-CNN: An Enhanced Explainable Convolutional Neural Network for Survival Prediction and Pathway Analysis in
Masrur Sobhan1, Md Mezbahul Islam1, Ananda Mohan Mondal1
1Knight Foundation School of Computing and Information Science, Florida International University, Miami, FL 33199, USA.
Biorxiv : the Preprint Server for Biology
|February 20, 2025
Summary
PathX-CNN improves glioblastoma multiforme (GBM) survival prediction by integrating multi-omics data. This explainable deep learning framework uses pathway-specific principal components for better accuracy and biological insights.
Area of Science:
- Bioinformatics
- Computational Biology
- Artificial Intelligence in Oncology
Background:
- Convolutional neural networks (CNNs) can analyze biological array data by converting it into image-like formats.
- Principal component analysis (PCA) captures global variance but struggles with sub-cohort specific variations, limiting CNN performance in cancer survival prediction.
- Existing methods for glioblastoma multiforme (GBM) survival prediction using CNNs often lack biological interpretability.
Purpose of the Study:
- To develop an explainable CNN framework, PathX-CNN, for improved multi-omics data integration and cancer survival prediction.
- To address the limitations of global variance capture in PCA for class-specific analyses.
- To enhance the biological interpretability of deep learning models in cancer research.
Main Methods:
- PathX-CNN integrates multi-omics data using pathway-based images derived from sub-cohort-specific principal components (PCs).
- The framework utilizes SHapley Additive exPlanations (SHAP) for explainable AI, identifying key pathways associated with survival.
- Performance was evaluated on glioblastoma multiforme (GBM) survival prediction and validated on other cancer types.
Main Results:
- PathX-CNN significantly outperformed existing pathway-based methods in predicting long-term survival (LTS) versus non-LTS in GBM.
- The SHAP analysis within PathX-CNN identified biologically plausible pathways linked to GBM survival.
- Experiments demonstrated PathX-CNN's superior performance over traditional approaches in diverse cancer types.
Conclusions:
- PathX-CNN offers a novel approach for integrating multi-omics data, enhancing prediction accuracy for cancer survival.
- The framework provides valuable pathway-specific insights into disease mechanisms, improving model interpretability.
- PathX-CNN highlights the potential of explainable CNNs for advancing precision oncology and understanding complex diseases.
