Harnessing computational intelligence for synthetic lethality: A roadmap from network biology to interpretable deep learning in precision oncology

Synthetic lethality (SL) is an emerging therapeutic paradigm in precision oncology that enables the selective targeting of cancer cells based on their specific genetic vulnerabilities. Despite its clinical promise, the enormous combinatorial search space of gene-gene interactions poses a major challenge for experimental screening. This review provides a critical survey of computational SL discovery, moving beyond sequential descriptions to offer a comparative synthesis of architectures ranging from traditional network-based models to advanced graph transformers and knowledge graph reasoning. We address a critical gap in the field by providing evidence-based best practices for methodological evaluation, including rigorous data splitting schemes, metric selection under class imbalance, and strategies for robust negative sampling. Furthermore, we discuss the clinical implications of synthetic rescue, a compensatory phenomenon that bypasses SL-induced cell death and acts as a primary mediator of drug resistance. By integrating multi-omics data, interpretable deep learning, and synthetic rescue mechanisms, we provide a practical translational framework for prioritizing candidates that are robust to resistance mechanisms. This roadmap aims to bridge the gap between computational prediction and actionable clinical insights in precision oncology.