Research map / Image generation / Autoregressive and latent generation
SpotDiff: Spatial Gene Expression Imputation Diffusion with Single-Cell RNA Sequencing Data Integration
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes SpotDiff, a multi-modal conditional diffusion model (DiT-based) that imputes spatial transcriptomics dropouts using spot-gene text prompts encoded by T5 and scRNA-seq data integrated per spot.
Problem
Spatial transcriptomics data are highly sparse due to low capture rates, and existing imputation methods only replicate observed distributions and underuse complementary scRNA-seq and multi-modal information.
Approach
Conditions a DiT-based diffusion model on (i) T5 embeddings of spot-gene textual descriptions via spot-gene cross-attention and (ii) scRNA-seq features aligned and fused per spot through a U-Net integration network with a spatially aware loss and cosine-similarity nearest-cell retrieval.
Key results
- On osmFISH, SpotDiff reaches PCC 0.3721 vs. 0.3020 for the second-best method SpaFormer.
- On FISH, PCC 0.6215 vs. 0.5876 for stMDCI.
- Ablation on osmFISH: adding scRNA-seq features f_rna raises PCC from 0.2987 baseline to 0.3612, and adding both f_rna and f_text reaches 0.3721.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use SpotDiff rather than SpaFormer | ST imputation · osmFISH | PCC: 0.3020 -> 0.3721 | Helped |
| use SpotDiff rather than stMDCI | ST imputation · FISH | PCC: 0.5876 -> 0.6215 | Helped |
| use SpotDiff rather than stMDCI | ST imputation · STARmap | PCC: 0.2874 -> 0.3215 | Helped |
| use SpotDiff rather than Tangram | ST imputation · MERFISH | PCC: 0.3421 -> 0.3784 | Helped |
| use SpotDiff rather than SpaFormer | ST imputation · 10x BA | PCC: 0.2401 -> 0.2559 | Helped |
| use SpotDiff rather than stMDCI | ST imputation · 10x HBC | PCC: 0.2244 -> 0.2345 | Helped |
| add frna while omitting ftext (ablation) | SpotDiff · osmFISH | PCC: 0.2987 -> 0.3215 | Helped |
| add ftext while omitting frna (ablation) | SpotDiff · osmFISH | PCC: 0.2987 -> 0.3612 | Helped |
| add frna and ftext (ablation) | SpotDiff · osmFISH | PCC: 0.2987 -> 0.3721 | Helped |
| add frna while omitting ftext (ablation) | SpotDiff · FISH | PCC: 0.5124 -> 0.5651 | Helped |
| add ftext while omitting frna (ablation) | SpotDiff · FISH | PCC: 0.5124 -> 0.6055 | Helped |
| add frna and ftext (ablation) | SpotDiff · FISH | PCC: 0.5124 -> 0.6215 | Helped |
Limitations
- Gains reported are measured against a limited set of baselines and metrics without variance or significance reporting in the text.
- Dependence on matched scRNA-seq data from the same tissue for integration.
Experimental setup
Datasets: osmFISH FISH STARmap MERFISH 10x BA 10x HBC Mop
Benchmarks: osmFISH FISH STARmap MERFISH 10x BA 10x HBC Mop
Metrics: PCC SSIM RMSE COSSIM ARI AMI NMI
Baselines: gimVI Tangram GraphST SpaFormer stMDCI stDiff STEM
Builds on: DiT denoising diffusion probabilistic models stDiff stMDCI
Closest papers by meaning
- Global Context-aware Representation Learning for Spatially Resolved Transcriptomics (ICML 2025 · similarity 0.86)
- SToFM: a Multi-scale Foundation Model for Spatial Transcriptomics (ICML 2025 · similarity 0.86)
- ASIGN: An Anatomy-aware Spatial Imputation Graphic Network for 3D Spatial Transcriptomics (CVPR 2025 · similarity 0.86)
- Gene-Gene Relationship Modeling Based on Genetic Evidence for Single-Cell RNA-Seq Data Imputation (NeurIPS 2024 · similarity 0.85)
- Diffusion Generative Modeling for Spatially Resolved Gene Expression Inference from Histology Images (ICLR 2025 · similarity 0.85)
- Scalable Generation of Spatial Transcriptomics from Histology Images via Whole-Slide Flow Matching (ICML 2025 · similarity 0.84)
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.