SHIFT: Survival Prediction from Incomplete and Heterogeneous Genomic Data

Short summary

Researchers introduce SHIFT, a Transformer-based survival prediction model that handles missing genomic data without test-time imputation using masked self-attention for robust multi-center deployment. Validated on glioblastoma and lung cancer across institutions with mismatched sequencing panels. Demonstrates incomplete patient cohorts can improve rather than degrade generalization.

• •SHIFT predicts patient survival from incomplete genomic data across heterogeneous sequencing panels using masked self-attention
• •Single model works with varying feature sets without imputation or cohort exclusion, enabling practical multi-center deployment
• •External validation on cancer survival shows strong cross-institutional generalization, suggesting precision oncology strategy

Generated with AI, which can make mistakes.