D3Impute: Dropout-aware discrimination, distribution-aware modeling, and density-guide imputation for scRNA-seq data
Siyi Huang1, Linfeng Jiang2, Ming Yi1
1School of Mathematics and Physics, China University of Geosciences, Wuhan, Hubei, China.
Plos Computational Biology
|December 1, 2025
Summary
D3Impute enhances single-cell RNA sequencing (scRNA-seq) analysis by accurately distinguishing technical zeros from biological zeros. This computational framework improves downstream analyses like cell clustering and gene expression detection.
Area of Science:
- Computational Biology
- Genomics
- Bioinformatics
Background:
- Single-cell RNA sequencing (scRNA-seq) reveals cellular heterogeneity but is challenged by technical "zeros" masking true gene expression.
- Distinguishing technical "zeros" from biological "zeros" is crucial for accurate data interpretation in scRNA-seq analysis.
- Current computational methods struggle to reliably differentiate these zero types, potentially distorting biological signals.
Purpose of the Study:
- To introduce D3Impute, a novel discriminative imputation framework designed to address the challenge of non-biological zeros in scRNA-seq data.
- To improve the accuracy of scRNA-seq data analysis by accurately identifying and handling technical zero measurements.
- To provide a robust and user-oriented solution for zero-inflated data in computational biology.
Main Methods:
- Developed D3Impute, a framework incorporating distribution-aware normalization, a dual-network discriminator utilizing bulk RNA-seq data, and a density-guided imputation engine.
- The dual-network discriminator leverages bulk RNA-seq as a reference to identify non-biological zeros while preserving biological zeros.
- The imputation engine recovers expression values while maintaining cellular neighborhood structures.
Main Results:
- D3Impute demonstrated significant improvements over 12 state-of-the-art methods across six diverse scRNA-seq datasets.
- The framework consistently enhanced downstream analyses, including cell clustering, trajectory inference, and differential expression detection.
- Evaluations confirmed D3Impute's robustness across varying data qualities, with guidelines for optimal application provided.
Conclusions:
- D3Impute offers a robust, biologically informed solution for handling non-biological zeros in scRNA-seq data.
- The framework significantly improves the accuracy and reliability of scRNA-seq data analysis.
- D3Impute provides a generalizable approach for managing zero-inflated data in computational biology.
