Explainable t-SNE for Single-Cell RNA-seq Data Analysis: A Cell-DrivenInformation Fusion Framework
- 1Baylor University
Abstract
Single-cell RNA sequencing fundamentally depends on defining cell-to-cell similarity to resolve distinct cell states. However, conventional dimensional-reduction frameworks such as t-SNE and UMAP inherently rely on Euclidean geometry, a construct ill-suited for high-dimensional, sparse count matrices, and produce cluster boundaries that lack biological interpretability. Here, we introduce cell-driven information fusion, an end-to-end framework that constructs cell affinity directly from raw counts by integrating three biological signals: gene detection patterns, expression magnitude of dominant transcripts, and global profile correlation. A sparsity-adaptive operator merges these signals into a unified affinity matrix, while an entry-usage diagnostic traces which biological component governs local cluster boundaries. The framework comprises two complementary implementations: c-TSNE for low-to-moderate sparsity (below 80 percent), which consistently outperforms Euclidean baselines across six benchmarks, and c-UMAP for high-sparsity regimes. Evaluated across 29 datasets and two protocols against 11 baseline methods, c-UMAP achieved the highest mean normalized and adjusted mutual information (NMI and AMI) in every setting, with all improvements remaining significant after Bonferroni correction. Because a single fixed weight vector was applied across all 29 datasets per protocol, these gains reflect intrinsic algorithmic robustness rather than dataset-specific hyperparameter tuning. Furthermore, c-UMAP scales seamlessly to 1.14 million cells, running up to 22-fold faster than the leading baseline. Crucially, by operating directly on raw counts without variance-based gene selection, our approach avoids the median 46.9 percent loss of top differentially expressed genes per cell class, preserving essential lineage markers such as IAPP and FOXP3. Together, these results demonstrate that atlas-scale efficiency, raw-data preservation, and mechanistic interpretability are fully compatible.
Related
- Explainable computational framework for subtype classification and biomarker identification in head and neck squamous cell carcinoma using machine learning
- ssJSD: A Fusion of Sparsity and Spatial Information for HiC Single-Cell Clustering
- Multi-model biological and sequence information fusion for gene regulatory network inference from single-cell transcriptomics
- Integrative Single-Cell Transcriptomics and Explainable AI for Disease Mechanism and Biomarker Discovery in Triple-Negative Breast Cancer
- A framework for designing splice-junction experiments in deep 3' single-cell RNA sequencing
BibTeX
@misc{han2026explainable,
title = {{Explainable t-SNE for Single-Cell RNA-seq Data Analysis: A Cell-DrivenInformation Fusion Framework}},
author = {H. Han and T. Zhang and L. Niu and J. Wang and F. Han and C. Li and J. Li and H. Cao},
howpublished = {bioRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.29.755455},
url = {https://www.biorxiv.org/content/10.64898/2026.09.29.755455v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Explainable t-SNE for Single-Cell RNA-seq Data Analysis: A Cell-DrivenInformation Fusion Framework
AU - H. Han
AU - T. Zhang
AU - L. Niu
AU - J. Wang
AU - F. Han
AU - C. Li
AU - J. Li
AU - H. Cao
PY - 2026
DA - 2026/09/29
PB - bioRxiv
DO - 10.64898/2026.09.29.755455
UR - https://www.biorxiv.org/content/10.64898/2026.09.29.755455v1
AB - Single-cell RNA sequencing fundamentally depends on defining cell-to-cell similarity to resolve distinct cell states. However, conventional dimensional-reduction frameworks such as t-SNE and UMAP inherently rely on Euclidean geometry, a construct ill-suited for high-dimensional, sparse count matrices, and produce cluster boundaries that lack biological interpretability. Here, we introduce cell-driven information fusion, an end-to-end framework that constructs cell affinity directly from raw counts by integrating three biological signals: gene detection patterns, expression magnitude of dominant transcripts, and global profile correlation. A sparsity-adaptive operator merges these signals into a unified affinity matrix, while an entry-usage diagnostic traces which biological component governs local cluster boundaries. The framework comprises two complementary implementations: c-TSNE for low-to-moderate sparsity (below 80 percent), which consistently outperforms Euclidean baselines across six benchmarks, and c-UMAP for high-sparsity regimes. Evaluated across 29 datasets and two protocols against 11 baseline methods, c-UMAP achieved the highest mean normalized and adjusted mutual information (NMI and AMI) in every setting, with all improvements remaining significant after Bonferroni correction. Because a single fixed weight vector was applied across all 29 datasets per protocol, these gains reflect intrinsic algorithmic robustness rather than dataset-specific hyperparameter tuning. Furthermore, c-UMAP scales seamlessly to 1.14 million cells, running up to 22-fold faster than the leading baseline. Crucially, by operating directly on raw counts without variance-based gene selection, our approach avoids the median 46.9 percent loss of top differentially expressed genes per cell class, preserving essential lineage markers such as IAPP and FOXP3. Together, these results demonstrate that atlas-scale efficiency, raw-data preservation, and mechanistic interpretability are fully compatible.
ER -