Single-Cell Feature Selection Method

Multi-omics Compatible

By fusing subspace distance and minimum redundancy, this method extends single-cell feature selection to multi-omics for the first time, selecting key genes across modalities for improved cell clustering.

Liu, Pei · Yuheng, Wang · Lv, Xiaoyi · Zhigang, Li · Chen, Cheng · Chen, Chen · Yuanzhi, He · Wang, Jing · Gu, Jin

Pattern Recognition 2026

Specifications

Number of evaluated datasets

{'zh': '16', 'en': '16'}

Advantages

Works across omics

The additive fusion structure allows separate processing of each omics before combination, enabling direct use on multi-omics feature selection.

Selects accurate genes

Variance–covariance subspace distance learns the subspace structure of the data, preserving genes most discriminative for clustering.

Removes redundancy

Inner product regularization and minimum redundancy terms are added to the objective function, directly removing redundant genes to produce a compact feature subset.

Applications

• Single-cell Multi-omics Analysis：Selects features across omics for better joint clustering
• Cell Clustering：Provides a compact and discriminative feature set for accurate clustering
• Trajectory Inference：Extracts genes important for cell state transitions to improve trajectory inference