Authors
Cen Bian, Ming Yang, Yuchen Han, Jing Yuan
Published in
IEEE transactions on computational biology and bioinformatics. Volume PP. Nov 12, 2026. Epub Nov 12, 2026.
Abstract
Cancer multi-omics data can characterize tumor heterogeneity from different molecular levels and provide an important information source for unsupervised cancer subtype discovery. However, existing tensorized multi-view subspace clustering methods still face problems in practical cancer sub typing tasks, including insufficiently compact tensor low-rank approximation, noise modeling limited to a single space, overly rigid constraints on cross-view relationships, and insufficient preservation of local boundary structures in the affinity matrix. To this end, this paper proposes a tensor subspace clustering method for cancer subtyping. This method stacks the clean self representation components of all views into a third-order tensor and introduces Schatten-p tensor low-rank regularization to more effectively preserve dominant singular structures. Meanwhile, a dual-space ℓ2,q joint noise model covering both data space and representation space is constructed to enhance robustness against heavy-tailed noise, heteroscedastic perturbations, and sample level outliers. Furthermore, this paper adopts HSIC to reduce cross-view covariance among the learned representations, and promotes a clear block-diagonal structure in the affinity matrix through sparse gradient graph regularization. Extensive evaluations on real-world cancer multi-omics datasets and standard machine learning datasets confirm its superior performance and consistent robustness for cancer subtype identification.
PMID:
43019266
Bibliographic data and abstract were imported from PubMed on 13 Nov 2026.
Read full publication at:
Please sign in
to see all details.
Advertisement
Stats
- Recommendations n/a n/a positive of 0 vote(s)
- Views 3
- Comments 0