Research map / Image generation / Fast diffusion sampling
ScDiVa: Masked Discrete Diffusion for Joint Modeling of Single-Cell Identity and Expression
Mingxuan Wang, Gaoyang Jiang, ZiJia Ren, Cheng Chen, Chuangxin Zhao, Lu Shi, Yanbiao Ma
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 scDiVa masked diffusion instead of GeneMamba for rank-value reconstruction | scDiVa · Immune reconstruction | Spearman: 0.844 -> 0.970 | Improved |
| Use scDiVa instead of GeneMamba on Pancreas reconstruction | scDiVa · Pancreas reconstruction | BLEU: 0.991 -> 0.965 | Hurt |
| Use multi-step diffusion (32 steps) instead of one-step denoising (ablation) | scDiVa · rank-value reconstruction | rank convergence / residual variance | Improved |
| Use scDiVa instead of recent foundation models for batch integration | scDiVa · Multi-batch integration (5 datasets) | Avg-batch / Avg-bio | Mixed |
| Use scDiVa instead of GeneMamba for fine-tuned annotation on imbalanced MS | scDiVa · MS annotation (fine-tuned) | Macro-F1: 0.5342 -> 0.7271 | Improved |
| Use scDiVa embeddings for zero-shot annotation versus scGPT and Geneformer | scDiVa · Zero-shot annotation (7 datasets) | accuracy: average accuracy 0.914 | Improved |
| Use scDiVa instead of GEARS/CellFM for perturbation prediction on Adamson | scDiVa · Adamson perturbation | C-DEP: 0.289 -> 0.337 (vs CellFM) | Improved |
| Use scDiVa instead of the Additive baseline on Norman double perturbations | scDiVa · Norman perturbation | C-DEP / HR@20: C-DEP 0.271 vs Additive 0.269; HR@20 0.441 vs 0.461 | Mixed |
Closest papers by meaning
- Learn from Your Mistakes: Self-Correcting Masked Diffusion Models (NeurIPS 2026 · similarity 0.87)
- Search-Augmented Masked Diffusion Models for Constrained Generation (NeurIPS 2026 · similarity 0.86)
- Why Masking Diffusion Works: Condition on the Jump Schedule for Improved Discrete Diffusion (NeurIPS 2025 · similarity 0.86)
- Simplified and Generalized Masked Diffusion for Discrete Data (NeurIPS 2024 · similarity 0.86)
- D-DOIT: Training-free Adaptation of Discrete Diffusion via Doob's h-Transform (NeurIPS 2026 · similarity 0.86)
- Beyond Masked and Unmasked: Discrete Diffusion Models via Partial Masking (NeurIPS 2025 · similarity 0.85)
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.
