Pathway-Attentive GAN for Interpretable Biomolecular Design

Abstract

High-throughput sequencing has greatly advanced cancer research, but a major gap remains in connecting TCGA transcriptomic data with detailed metabolomic profiles. This disconnect limits our understanding of metabolic changes that drive tumor progression and resistance to treatment. To address this, we introduce the Pathway-Attentive GAN (PathGAN), a new framework that combines transformer-based attention mechanisms with a GNN discriminator to generate realistic and biologically relevant metabolite profiles as a case study. We validate these profiles using COBRApy-based flux balance analysis to ensure they align with key metabolic pathways. By linking transcriptomics and metabolomics, PathGAN improves our understanding of tumor metabolism and provides valuable insights for cancer therapy. We believe this work can offer a powerful tool for precision oncology, helping to develop more targeted and effective treatments.