Table 4. Predictions of multi-omics molecular features of HR+/HER2− breast cancer based on pathological whole slide images using deep learning.
| Prediction categories | Multi-omics molecular features | Validation AUC (95% CI) |
Test AUC (95% CI) |
|---|---|---|---|
| Somatic mutations | PIK3CA | 0.68 (0.56–0.80) | 0.58 (0.44–0.71) |
| TP53 | 0.73 (0.62–0.85) | 0.68 (0.56–0.81) | |
| GATA3 | 0.50 (0.31–0.68) | 0.68 (0.47–0.89) | |
| MAP3K1 | 0.81 (0.63–0.99) | 0.42 (0.22–0.61) | |
| KMT2C | 0.69 (0.49–0.89) | 0.57 (0.36–0.79) | |
| AKT1 | 0.72 (0.49–0.94) | 0.49 (0.27–0.71) | |
| PTEN | 0.85 (0.72–0.97) | 0.62 (0.40–0.83) | |
| FAT3 | 0.84 (0.67–1.00) | 0.61 (0.16–1.00) | |
| SF3B1 | 0.75 (0.49–1.00) | 0.47 (0.00–1.00) | |
| Gene set enrichment analysis scores | Angiogenesis | 0.63 (0.51–0.75) | 0.52 (0.39–0.65) |
| DNA repair | 0.70 (0.59–0.82) | 0.63 (0.50–0.75) | |
| Estrogen response early | 0.64 (0.50–0.77) | 0.48 (0.34–0.62) | |
| G2-M checkpoint | 0.87 (0.80–0.95) | 0.79 (0.69–0.90) | |
| IFN-γ response | 0.72 (0.61–0.84) | 0.62 (0.50–0.75) | |
| PI3K/AKT/mTOR signaling | 0.71 (0.60–0.82) | 0.63 (0.50–0.76) | |
| Immunotherapy biomarkers | iTILs | 0.59 (0.35–0.82) | 0.78 (0.55–1.00) |
| sTILs | 0.76 (0.65–0.87) | 0.76 (0.65–0.87) | |
| CD8A mRNA | 0.66 (0.54–0.77) | 0.71 (0.60–0.82) | |
| PDCD1 mRNA | 0.68 (0.56–0.79) | 0.74 (0.63–0.85) | |
| CD274 mRNA | 0.64 (0.52–0.76) | 0.51 (0.39–0.64) |
AUC, area under the curve; CI, confidence interval; iTILs, intratumoral tumor-infiltrating lymphocytes; sTILs, stromal tumor-infiltrating lymphocytes.