A multi-omics signature to predict the prognosis of invasive ductal carcinoma of the breast

Abstract

Background: Precisely evaluating the prognosis of invasive ductal carcinoma (IDC) of the breast is challenging as most prognostic signatures use single-omics data based on gene or clinical information.

Methods: Whole-slide images (WSIs), transcriptome, and clinical data of breast IDC were collected from the Cancer Genome Atlas Database. The cancer-associated fibroblast (CAF) gene sets were downloaded from the Molecular Signatures Database. The WSI feature was extracted by artificial feature engineering. The CAF prognostic genes were determined by the Gene Set Enrichment Analysis, the Wilcoxon test, and univariate Cox regression. The IDC patients were divided into the training and test sets. The prognostic signatures based on WSIs, IDC-CAFs, bi-omics, and tri-omics were constructed using multivariate Cox regression. The samples were divided into low- and high-risk groups according to the median risk score. The Kaplan-Meier survival and receiver operating characteristic curves were applied to validate the prediction performance of the four signatures.

Results: In total, 508 IDC patients with complete data were included. The area under the curve (AUC) of single-omics signature based on WSI characteristics and CAFs was 0.765 and 0.775, whereas the AUC of bi-omics was 0.823. The tri-omics signature based on WSIs, CAFs, and lymph node status demonstrated the best predictive value with an AUC of 0.897.

Conclusion: The multi-omics signature based on WSIs, CAFs, and clinical characteristics showed excellent prediction ability in breast IDC patients, whose risk factors can also provide a valuable diagnostic reference for the clinical course.

Keywords: Cancer-associated fibroblasts; Invasive ductal carcinoma; Prognosis; Tri-omics; Whole-slide images.