Imputation for transcription factor binding predictions based on deep learning
Problem
ChIP-seq data covers only a small fraction of possible transcription factor–cell line combinations, leaving most cell-specific TF binding patterns unknown.
Why it matters
Mapping gene regulatory networks is fundamental to understanding biological systems and disease.
What the AI did
TFImpute, a multi-task deep neural network, predicts cell-specific TF binding for untested TF-cell line combinations using only 4% of the combinations.
Results
Comparable accuracy to existing methods on TF-cell line combinations with ChIP-seq data, using 4% of combinations; abstract truncated before further results.
Impact
Enables prediction of TF binding without exhaustive ChIP-seq experiments, per the authors' claims.
Record
| Venue | PLoS Computational Biology |
|---|---|
| Published | 2017 |
| Work type | article |
| Authors | Qian Qin, Jianxing Feng |
| Citations | 96 (OpenAlex, at import time) |
| References | 46 |
| Model or method | TFImpute (deep neural network, multi-task learning) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1371/journal.pcbi.1005403 |
| Open access | gold |
| Keywords | transcription factors, ChIP-seq, deep learning, multi-task learning, binding prediction |
Importance rating
real but narrower advance
Mapping gene regulatory networks is called fundamental, but no scale stated; a computational tool reducing need for exhaustive ChIP-seq experiments.