PATH: A New Gene Embedding Strategy for Cancer Prognosis Prediction

A novel strategy known as PATH tackles the challenges of predicting cancer progression by implementing a modulation-based, patient-specific gene embedding technique. This innovation marks a significant departure from traditional hierarchical models and integration frameworks, which frequently struggle to learn distinct base representations for individual genes. The complexity of molecular omics data across patients complicates accurate cancer progression predictions. Although biologically informed models have enhanced interpretability, they still face challenges in encoding individual genes for pathway representation. Current hierarchical models extract gene features by directly mapping raw molecular data, while integration frameworks often depend on basic statistical aggregations of patient signals. PATH initiates with a unified base embedding for each gene, maintaining a consistent biological identity. This research was shared on arXiv under identifier 2604.16685v1 as a cross announcement.