Federated and Differentially Private Prediction of Immune Checkpoint Blockade Response from Multi-Cohort Transcriptomics
Abstract
Transcriptomic signatures of immune checkpoint blockade (ICB) response often exhibit limited transportability across independent cohorts, while privacy constraints restrict pooling of pa-tient-level transcriptomic data across institutions. We evaluated whether federated learning (FL) with record-level differential privacy (DP) could predict ICB response across heterogeneous cohorts and whether a pre-specified PD-L1-trafficking module added predictive signal beyond an inter-feron-γ (IFN-γ) signature. Five public ICB cohorts (four anti-PD1 melanoma and one anti-PD-L1 urothelial cohort) were treated as federated sites. Logistic regression was trained using FedAvg and FedProx with per-example DP-SGD and cumulative Rényi DP accounting. The evaluation was conducted through leave-one-cohort-out external validation against a matched centralized non-private reference. The mean area under the receiver operating characteristic curves (AUROCs) were modest, ranging from approximately 0.56 to 0.61, and remained broadly stable across the privacy budgets evaluated. Specifically, for the combined feature set, mean AUROC was 0.561 without DP and 0.573 at ε = 1. The IFN-γ module provided the strongest predictive signal (≈0.61), whereas trafficking alone performed near chance (≈0.49), and combining the modules did not im-prove overall discrimination (≈0.56). These results support the feasibility of privacy-preserving federated ICB-response prediction despite significant cross-cohort…
Read the paper
Related studies
- Active Learning during Federated Learning for Object Detection
- Fairness-aware federated learning
- Decentralized Federated Learning
- Building a Personally Identifiable Information Recognizer in a Privacy Preserved Manner Using Automated Annotation and Federated Learning
- Preservation of Patient Level Privacy: Federated Classification and Calibration Models
- Efficient explainable federated learning
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.