DISC: a highly scalable and accurate inference of gene expression and structure for single-cell transcriptomes using semi-supervised deep learning
Problem
Dropouts distort gene expression and misclassify cell types in single-cell transcriptome data.
Why it matters
Dropout-induced errors undermine gene expression and cell type identification in sparse scRNA-seq analyses.
What the AI did
DISC, a semi-supervised deep learning network, infers gene structure and expression obscured by dropouts to improve downstream analysis.
Results
Consistently outperformed seven state-of-the-art imputation approaches across ten real-world datasets (no numeric metrics given).
Impact
DISC is presented as a scalable, reliable tool for recovering gene expression and improving cell type identification.
Record
| Venue | Genome biology |
|---|---|
| Published | 2020 |
| Work type | article |
| Authors | Yao He, Hao Yuan, Cheng Feng Wu, Zhi Xie |
| Citations | 36 (OpenAlex, at import time) |
| References | 45 |
| Model or method | DISC (semi-supervised deep learning network) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1186/s13059-020-02083-3 |
| Open access | gold |
| Keywords | deep learning, single-cell RNA-seq, dropout imputation, semi-supervised learning, cell type identification |
Importance rating
real but narrower advance
Dropout distortion affects sparse scRNA-seq analyses broadly, but abstract gives no scale or numbers; a useful methodological advance for a widely used field.