Related Experiment Video
Updated: Jun 12, 2026
13:24
Integration of Wet and Dry Bench Processes Optimizes Targeted Next-generation Sequencing of Low-quality and Low-quantity Tumor Biopsies
Published on: April 11, 2016
11.8K
DeepDeconUQ estimates malignant cell fraction prediction intervals in bulk RNA-seq tissue
Jiawei Huang1, Yuxuan Du1,2, Kevin R Kelly3
1Department of Quantitative and Computational Biology, University of Southern California, Los Angeles, California, United States of America.
Plos Computational Biology
|June 4, 2025
Summary
DeepDeconUQ quantifies uncertainty in malignant cell fraction estimation using bulk RNA-seq data. This deep learning model provides reliable prediction intervals, improving cancer diagnosis and research accuracy.
Area of Science:
- Computational Biology
- Genomics
- Bioinformatics
Background:
- Accurate malignant cell fraction estimation is crucial for cancer diagnosis, prognosis, and treatment.
- Current methods often lack uncertainty quantification, limiting their clinical and research utility.
Purpose of the Study:
- To introduce DeepDeconUQ, a deep neural network model for estimating prediction intervals of malignant cell fractions from bulk RNA-seq data.
- To integrate uncertainty quantification into cancer cell fraction predictions.
Main Methods:
- DeepDeconUQ utilizes single-cell RNA sequencing (scRNA-seq) data and conformalized quantile regression.
- A quantile regression neural network establishes prediction interval bounds, followed by a calibration step for statistical validity and discrimination.
Main Results:
- DeepDeconUQ outperforms existing methods in coverage accuracy and interval tightness on simulated and real cancer datasets.
- The model demonstrates robustness against gene expression perturbations.
Conclusions:
- DeepDeconUQ offers a robust approach for quantifying uncertainty in malignant cell fraction estimation.
- This method enhances the reliability of cancer cell fraction predictions for clinical and research applications.
