Imputation of single‐cell gene expression with an autoencoder neural network
Problem
High dropout rates in single-cell RNA-seq produce many false zero gene counts that complicate downstream analyses.
Why it matters
Dropout noise can distort biological conclusions drawn from single-cell gene expression data.
What the AI did
The LATE method trains an autoencoder to treat zeros as missing values and impute them; TRANSLATE adds transfer learning from a reference gene expression dataset for initialization.
Results
On both simulated and real data, LATE and TRANSLATE outperform existing scRNA-seq imputation methods (metrics truncated in abstract)
Impact
The paper provides AI-based imputation tools that outperform existing methods on simulated and real data.
Record
| Venue | Quantitative Biology |
|---|---|
| Published | 2020 |
| Work type | article |
| Authors | Md. Bahadur Badsha, Rui Li, Liu Boxiang, Yang I. Li, Min Xian, Nicholas E. Banovich, Audrey Qiuyan Fu |
| Citations | 56 (OpenAlex, at import time) |
| References | 24 |
| Model or method | LATE and TRANSLATE (autoencoder neural networks) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1007/s40484-019-0192-7 |
| Open access | diamond |
| Keywords | single-cell RNA-seq, imputation, autoencoder, deep learning, dropouts, transfer learning |
Importance rating
real but narrower advance
Dropout noise affects scRNA-seq analyses broadly; abstract notes explosive growth of the technology but frames this as a methods limitation, a real but narrower computational advance.