Cell-JEPA: Latent Representation Learning for Single-Cell Transcriptomics
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866912868424744960 |
|---|---|
| author | ElSheikh, Ali Wang, Rui-Xi Wu, Weimin Wen, Yibo Dibaeinia, Payam Zhang, Jennifer Yuntong Hu, Jerry Yao-Chieh Knudson, Mei Babu, Sudarshan Sun, Shao-Hua Khan, Aly A. Liu, Han |
| author_facet | ElSheikh, Ali Wang, Rui-Xi Wu, Weimin Wen, Yibo Dibaeinia, Payam Zhang, Jennifer Yuntong Hu, Jerry Yao-Chieh Knudson, Mei Babu, Sudarshan Sun, Shao-Hua Khan, Aly A. Liu, Han |
| contents | Single-cell foundation models learn by reconstructing masked gene expression, implicitly treating technical noise as signal. With dropout rates exceeding 90%, reconstruction objectives encourage models to encode measurement artifacts rather than stable cellular programs. We introduce Cell-JEPA, a joint-embedding predictive architecture that shifts learning from reconstructing sparse counts to predicting in latent space. The key insight is that cell identity is redundantly encoded across genes. We show predicting cell-level embeddings from partial observations forces the model to learn dropout-robust features. On cell-type clustering, Cell-JEPA achieves 0.72 AvgBIO in zero-shot transfer versus 0.53 for scGPT, a 36% relative improvement. On perturbation prediction within a single cell line, Cell-JEPA improves absolute-state reconstruction but not effect-size estimation, suggesting that representation learning and perturbation modeling address complementary aspects of cellular prediction. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_02093 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Cell-JEPA: Latent Representation Learning for Single-Cell Transcriptomics ElSheikh, Ali Wang, Rui-Xi Wu, Weimin Wen, Yibo Dibaeinia, Payam Zhang, Jennifer Yuntong Hu, Jerry Yao-Chieh Knudson, Mei Babu, Sudarshan Sun, Shao-Hua Khan, Aly A. Liu, Han Computational Engineering, Finance, and Science Single-cell foundation models learn by reconstructing masked gene expression, implicitly treating technical noise as signal. With dropout rates exceeding 90%, reconstruction objectives encourage models to encode measurement artifacts rather than stable cellular programs. We introduce Cell-JEPA, a joint-embedding predictive architecture that shifts learning from reconstructing sparse counts to predicting in latent space. The key insight is that cell identity is redundantly encoded across genes. We show predicting cell-level embeddings from partial observations forces the model to learn dropout-robust features. On cell-type clustering, Cell-JEPA achieves 0.72 AvgBIO in zero-shot transfer versus 0.53 for scGPT, a 36% relative improvement. On perturbation prediction within a single cell line, Cell-JEPA improves absolute-state reconstruction but not effect-size estimation, suggesting that representation learning and perturbation modeling address complementary aspects of cellular prediction. |
| title | Cell-JEPA: Latent Representation Learning for Single-Cell Transcriptomics |
| topic | Computational Engineering, Finance, and Science |
| url | https://arxiv.org/abs/2602.02093 |