Research map / Image generation / Fast diffusion sampling
Adapting a Pre-trained Single-Cell Foundation Model to Spatial Gene Expression Generation from Histology Images
Donghai Fang, Yongheng Li, Zhen Wang, Yuansong Zeng, Wenwen Min
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use pretrained CellFM instead of random init (ablation) | HINGE · cSCC P2 ST rep3 slice (ablation) | PCC-50: 0.7425 -> 0.8755 | Improved |
| keep CellFM frozen instead of tuning decoder (ablation) | HINGE · cSCC P2 ST rep3 slice (ablation) | PCC-50: 0.8518 -> 0.8755 | Improved |
| keep CellFM frozen instead of LoRA on backbone (ablation) | HINGE · cSCC P2 ST rep3 slice (ablation) | PCC-50: 0.8457 -> 0.8755 | Improved |
| use masked diffusion instead of Gaussian diffusion (ablation) | HINGE · cSCC P2 ST rep3 slice (ablation) | PCC-50: 0.7738 -> 0.8755 | Improved |
| add warm-start curriculum (ablation) | HINGE · cSCC P2 ST rep3 slice (ablation) | PCC-50: 0.8691 -> 0.8755 | Improved |
| fill masked entries with random instead of zero values (ablation) | HINGE · cSCC P2 ST rep3 slice (ablation) | PCC-50: 0.8752 vs 0.8755 | No clear effect |
| use SoftAdaLN instead of histology-conditioned affine LN (ablation) | HINGE · cSCC P2 ST rep3 slice (ablation) | PCC-50: 0.8298 -> 0.8755 | Improved |
| add SoftNorm to SoftAdaLN (ablation) | HINGE · cSCC P2 ST rep3 slice (ablation) | PCC-50: 0.8658 -> 0.8755 | Improved |
| add identity initialization to SoftAdaLN (ablation) | HINGE · cSCC P2 ST rep3 slice (ablation) | PCC-50: 0.8631 -> 0.8755 | Improved |
| concatenate UNI and CONCH encoders instead of UNI alone (ablation) | HINGE · cSCC P2 ST rep3 slice (ablation) | PCC-50: 0.8625 -> 0.8755 | Improved |
| HINGE versus best baseline on Kidney error metrics | HINGE · Kidney | MSE: 1.315 (TRIPLEX) vs 1.459 | Hurt |
| HINGE versus STFlow | HINGE · cSCC | PCC-50: 0.678 -> 0.705 | Improved |
Closest papers by meaning
- Towards Scalable Context-Aware Single-Cell Spatial Transcriptomics Prediction from Histology Images (NeurIPS 2026 · similarity 0.93)
- Diffusion Generative Modeling for Spatially Resolved Gene Expression Inference from Histology Images (ICLR 2025 · similarity 0.93)
- Scalable Generation of Spatial Transcriptomics from Histology Images via Whole-Slide Flow Matching (ICML 2025 · similarity 0.91)
- Preserving DEG Rankings for Gene Discovery in Histology-Based Spatial Gene Expression Prediction (NeurIPS 2026 · similarity 0.90)
- HEST-1k: A Dataset For Spatial Transcriptomics and Histology Image Analysis (NeurIPS 2024 · similarity 0.89)
- GeneFlow: Translation of Single-cell Gene Expression to Histopathological Images via Rectified Flow (NeurIPS 2025 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
