Research map / Probabilistic and scientific modeling / Molecules and proteins
SAVE: A Generalizable Framework for Multi-Condition Single-Cell Generation with Gene Block Attention
Jiahao Li, Jiayi Dong, Peng Ye, Xiaochi Zhou, Haohai Lu, Fei Wang
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 |
|---|---|---|---|
| add gene block attention (ablation) | SAVE · Heart (dual-condition generation) | Wasserstein distance: 8.89 -> 8.30 | Improved |
| add gene block attention (ablation) | SAVE · Heart (dual-condition generation) | MMD: 0.65 -> 0.63 | Improved |
| increase gene block size K from 600 to 3200 (ablation) | SAVE · Heart (dual-condition generation) | Wasserstein distance: 9.70 -> 8.30 | Improved |
| increase gene block size K beyond 3200 (4000, 5600) (ablation) | SAVE · Heart (dual-condition generation) | Wasserstein distance: 8.30 -> 8.37 / 8.41 | No clear effect |
| use gene blocks (K=3200) instead of per-gene attention (K=1) (ablation) | SAVE · training time under 24GB VRAM | training time (min): 2391.2 -> 12.5 min | Improved |
| add condition masking (ratio 0.6) (ablation) | SAVE · PBMC-IFN perturbation prediction | average PCC / R2: 0.90 -> 0.96 PCC; 0.75 -> 0.86 R2 | Improved |
| use SAVE instead of CFGen on single-condition generation | SAVE · PBMC3K and Tabula Muris | Wasserstein distance: 16.94 vs 19.14 (PBMC3K); 7.39 vs 10.86 (Tabula Muris) | Hurt |
| use SAVE instead of baselines on Dentate gyrus | SAVE · Dentate gyrus | Wasserstein distance: 21.55 -> 9.16 | Improved |
| use SAVE for dual-condition generation vs best baseline | SAVE · Heart, PBMC, Lung Atlas | Wasserstein distance: 12.57 -> 8.30; 11.38 -> 5.37; 13.89 -> 4.37 | Improved |
| use SAVE on held-out Lung Cancer conditions | SAVE · Lung Cancer (unseen) | WD / MMD: WD 5.29 -> 4.63; MMD 1.87 vs 2.11 | Mixed |
| use SAVE for perturbation prediction vs strongest baselines | SAVE · PBMC-IFN | average PCC: 0.94 -> 0.96 | Improved |
Closest papers by meaning
- MUSS: A Multi-scale and Sequence-based Model for Single-cell Gene Regulation (NeurIPS 2026 · similarity 0.86)
- Multi-Modal and Multi-Attribute Generation of Single Cells with CFGen (ICLR 2025 · similarity 0.86)
- STRAND: Sequence-Conditioned Transport for Single-Cell Perturbations (NeurIPS 2026 · similarity 0.84)
- ViTally Consistent: Scaling Biological Representation Learning for Cell Microscopy (ICML 2025 · similarity 0.84)
- sciLaMA: A Single-Cell Representation Learning Framework to Leverage Prior Knowledge from Large Language Models (ICML 2025 · similarity 0.84)
- A Simple and Comprehensive Benchmark for Single-Cell Transcriptomics (AAAI 2025 · similarity 0.83)
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.
