genomic-data

SHIFT is a missingness-aware survival model that uses masked self-attention to predict from incomplete genomic inputs without test-time imputation, showing strong generalization across cohorts in glioblastoma and lung squamous cell carcinoma.