Layer-Specific Source-Anchored Transfer Learning for Multi-Omics Survival Prediction in Younger Breast Cancer Patients

LSAT-EN improves 2–4-year breast cancer survival prediction for patients aged <50, reaching AUCs up to 0.789 under high source-target heterogeneity (sigma^2=4).

• Why it matters: Younger patients are underrepresented, and survival models trained on older cohorts can lose accuracy when source and target differ.
• What they did: The authors used layer-specific source-anchored transfer learning (LSAT-EN) in group-correlated simulations and in TCGA-BRCA integrating 6 omics layers, evaluating AUC, Harrell’s C-index, and IPCW AUC.
• The result: In TCGA-BRCA, LSAT-EN improved discrimination versus overall and individual EN across binary and Cox models for all time horizons and source age groups, and it highlighted mRNA/miRNA biomarkers and miRNA–mRNA pairs.

AI summary of the abstract · may contain errors · not medical advice · read the paper.