Views
No views yet
DELPHI: Dual-scale calibrated confidence enables trustworthy spatial gene expression prediction from histology. Under review.
| File | Size | Description |
|---|---|---|
delphi_her2st.pt | 34 MB | HER2ST full-dataset model (seed 42, epoch 22) |
1from src.model import DELPHI
2import torch
3
4model = DELPHI(uni2h_dim=1536, hidden_dim=384, num_genes=785,
5 gh=12, gw=12, knn_k=8, n_swin_blocks=4)
6model.load_state_dict(torch.load("delphi_her2st.pt", map_location="cpu"))
7model.eval()
8
9# Inference: x [N, 1536] UNI2-h features, pos [N, 2] coordinates
10with torch.no_grad():
11 mu, log_phi, pi, sigma_al, sigma_ep = model(x, pos)1@article{delphi2025,
2 title={DELPHI: Dual-scale calibrated confidence enables trustworthy
3 spatial gene expression prediction from histology},
4 author={...},
5 journal={Under review},
6 year={2025}
7}