Context-Aware Dual-Graph Neural Network for Cellular State Inference under Genetic Perturbations

Mapping transcriptional responses to genetic perturbations is essential across diverse biomedical contexts, ranging from understanding genetic interactions in oncology to advancing novel regenerative therapies. Despite advances in experimental tools, the combinatorial complexity of potential perturbations renders exhaustive wet-lab screening prohibitively costly and time-consuming. Meanwhile, existing high-throughput methods remain limited by incomplete regulatory networks and linear interaction assumptions. Emerging deep learning approaches often require prior experimental data for every perturbation to predict combinatorial effects. In this work, we introduce the context-aware dual-graph inference neural network (ADGI) to predict transcriptional outcomes under genetic perturbations. ADGI integrates prior knowledge with context-aware mechanisms to model regulatory dynamics, thereby enabling the prediction of transcriptional responses to novel perturbations and their combinations. Additionally, we developed a perturbation scheduling strategy that optimises training efficiency and guides targeted inference. Experimental results demonstrate that the proposed method achieves promising performance compared to established models. © 2026 IEEE.