PASNet

PASNet predicts patient prognosis from high-dimensional genomic data by integrating biological pathways into a sparse deep neural network to model nonlinear and hierarchical gene–pathway relationships for survival prediction.

Key Features:

• Pathway-Based Analysis: Incorporates biological pathways into the predictive model to capture hierarchical relationships between genes and pathways.
• Sparse Deep Neural Network Architecture: Employs a sparse deep neural network structure to focus model capacity on a subset of pathway-associated nodes.
• Model Interpretability: Produces a sparse solution that highlights key biological components contributing to prognostic predictions.
• High-Dimensional, Low-Sample-Size Handling: Tailored to handle the complexity and dimensionality challenges of high-dimensional and low-sample-size genomic datasets.
• Robust Predictive Performance in GBM: Demonstrated higher Area Under the Curve (AUC) and F1-scores on Glioblastoma multiforme (GBM) compared with existing classifiers.
• Cross-Validation Assessment: Performance was evaluated through multiple cross-validation experiments.
• Statistical Validation: Enhanced performance significance was confirmed using the Wilcoxon signed-rank test.
• Biological Relevance: Identified pathways align with pathways recognized in prior GBM biological and medical research.

Scientific Applications:

• Prognosis Prediction in Cancer: Predicts long-term survival outcomes in cancers, including Glioblastoma multiforme (GBM).
• Gene–Pathway Relationship Modeling: Models multilayered biological systems to elucidate intricate relationships between genes and pathways in genomic medicine.
• Hypothesis Generation and Validation: Facilitates identification of pathway-level candidates for downstream biological hypothesis testing and therapeutic investigation.

Methodology:

Integrates biological pathways into a sparse deep neural network that models nonlinear effects and hierarchical representations of genes and pathways, with performance evaluated by multiple cross-validation experiments and significance tested by the Wilcoxon signed-rank test.