ScGeo reveals non-canonical trajectories beyond RNA velocity in radiation-induced hematopoietic recovery.

Authors

Yu-Chen Liu, Kengo Yoshida

Published in

Bioinformatics advances. Volume 6. Issue 1. Pages vbag266. Epub Sep 09, 2026.

Abstract

Low-dimensional representations are central to single-cell RNA sequencing analysis, yet perturbation-associated geometry is often interpreted visually without explicit assessment of estimator, sampling, or representation dependence. We introduce ScGeo, a representation-aware framework that treats embeddings as quantitative objects and reports robust state displacement, biological-sample uncertainty, local geometric preservation, cross-representation stability, distributional change, and agreement between condition-dependent displacement and independently supplied dynamics estimates. In GSE280305 post-irradiation hematopoietic recovery, ScGeo identified heterogeneous D8-to-D21 cluster displacement and partial agreement between geometric shifts and RNA velocity, while avoiding interpretation of time-point mixing as proof of valid integration. A prespecified synthetic benchmark showed that robust center estimators reduced outlier sensitivity, global representation corruption was detectable, and fine-grained localization of local distortion remained limited. A GSE132188-derived pancreatic-development workflow provided descriptive geometry-dynamics validation. In GSE249479, inflammatory effects in hematopoietic stem and progenitor cells were broadly stable across the primary representation ensemble but remained descriptive because biological-replicate identity was unavailable. In replicate-aware GSE211713 lung-radiation analysis, early 17 Gy effects were representation-sensitive, whereas late remodeling was stable in five of six major compartments. ScGeo provides an auditable downstream layer for distinguishing stable, neutral, insufficient-coverage, and representation-sensitive interpretations rather than assuming any single latent space is biologically definitive.

PMID:
42801126
Bibliographic data and abstract were imported from PubMed on 27 Sep 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.