Synthetic-Lethality Prediction via STATE-Based Multi-Omics Modeling
Predicting synthetic-lethal gene pairs for cancer-selective drug-target prioritization by integrating Perturb-seq perturbation transcriptomics, DepMap functional dependency (CRISPR GeneEffect), and ESM2 protein-sequence embeddings. Models perturbation responses as cell-state representations via a prototype-based distribution-regression approach that recovers dependency signal missed by pseudobulk and attention baselines.