Liu, Pei, Wang, Yuheng, Lv, Xiaoyi, Li, Zhigang, Chen, Cheng, Chen, Chen, He, Yuanzhi  ORCID: https://orcid.org/0009-0007-8424-2654, Wang, Jing and Gu, Jin
2026.
A single-cell feature selection method based on subspace and minimum redundancy, applicable to multi-omics.
Pattern Recognition
180
(Part B)
, 114094.
10.1016/j.patcog.2026.114094
|
Preview |
PDF
- Accepted Post-Print Version
Available under License Creative Commons Attribution. Download (35MB) | Preview |
Abstract
Feature selection is a crucial step in improving the reliability and accuracy of single-cell clustering. Existing feature selection methods exhibit strong modality dependence, neglect relevant subspace structures, and cannot be extended to single-cell multi-omics applications. Therefore, this paper proposes for the first time a single-cell clustering feature selection method based on the fusion of subspace distance and minimum redundancy (sc-MSDMR). This method selects important features for accurate single-cell clustering and can be applied to single-cell multi-omics. First, sc-MSDMR constructs an objective function using variance–covariance subspace distance to perform subspace learning on the feature and statistical information of gene data. Then, inner product regularization and minimum redundancy terms are added to the objective function to remove redundant genes and select significant genes. Simultaneously, an efficient optimization algorithm is proposed to continuously update the objective function for sc-MSDMR until it converges. Finally, validation and comparison experiments are conducted on 16 single-cell RNA sequencing datasets, demonstrating that sc-MSDMR outperforms other state-of-the-art feature selection methods. Meanwhile, differentially expressed gene analysis validated its biological interpretability and practicality. Furthermore, due to its additive fusion structure and cross-modal commonality, sc-MSDMR is suitable for multi-omics tasks. Applying sc-MSDMR to other omics cell trajectory inference tasks and multi-omics cell clustering tasks demonstrates excellent multi-omics application capabilities and scalability.
| Item Type: | Article |
|---|---|
| Date Type: | Publication |
| Status: | Published |
| Schools: | Schools > Computer Science & Informatics |
| Additional Information: | RRS policy applied. License information from Publisher: LICENSE 1: Title: This article is under embargo with an end date yet to be finalised. |
| Publisher: | Elsevier |
| ISSN: | 0031-3203 |
| Date of First Compliant Deposit: | 7 July 2026 |
| Date of Acceptance: | 24 May 2026 |
| Last Modified: | 05 Aug 2026 13:00 |
| URI: | https://orca.cardiff.ac.uk/id/eprint/187387 |
Actions (repository staff only)
 |
Edit Item |
Dimensions
Dimensions