Dimensionality Reduction for Cyclic Compositional Data
- 1California Institute of Technology
Abstract
The cyclic nature of many biological systems is frequently reflected in periodic gene expression patterns. Standard dimensionality reductions can reveal such variation but do not link a sample's position on a cycle to an ordering of the molecular features that define it. We introduce clrcycle, which applies a centered log-ratio transformation, learns a cyclic feature ordering by maximizing variance captured by the first Fourier mode, and projects samples onto the corresponding cosine and sine coordinates. The resulting two-dimensional geometry assigns each sample a phase and a radius measuring the magnitude of its projection onto the learned Fourier circle. Across a mouse circadian atlas, clrcycle recovers known temporal organization and reveals tissue-specific differences in rhythmic timing. In FUCCI-profiled single cells, clrcycle represents independently measured cell-cycle phase in two-dimensional coordinates more reliably than conventional two-dimensional embeddings. In human endometrium, a circular coordinate learned from 250 highly variable genes recovers menstrual-stage organization and can stage unseen endometrial biopsies. Together, these results establish clrcycle as powerful, explicitly circular dimensionality reduction technique to quantify high-dimensional cyclic variation across biological cycles.
Related
- Evaluating the Robustness of Path-Preservation Benchmarks for Dimensionality Reduction Across Point-Density Thresholds: Linear and Cyclic Single-Cell Trajectories
- multiTEMPTED: Joint Dimensionality Reduction of Longitudinal Multi-omic Data with Modality-Specific Temporal Dynamics
- Intrinsic dimensionality of deep learning representations reveals cell death-associated heterogeneity in Parkinsons disease iPSC-derived neurons
- High-dimensionality approaches to the understanding and therapeutic manipulation of autoimmunity
- Dimensionality-induced critical phase transition in stochastic Lotka-Volterra equation: From statistical averaging to systemic tipping point
BibTeX
@misc{swarna2026dimensionality,
title = {{Dimensionality Reduction for Cyclic Compositional Data}},
author = {N. P. Swarna and J. Emhemed and L. Pachter},
howpublished = {bioRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.22.761053},
url = {https://www.biorxiv.org/content/10.64898/2026.10.22.761053v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Dimensionality Reduction for Cyclic Compositional Data
AU - N. P. Swarna
AU - J. Emhemed
AU - L. Pachter
PY - 2026
DA - 2026/10/22
PB - bioRxiv
DO - 10.64898/2026.10.22.761053
UR - https://www.biorxiv.org/content/10.64898/2026.10.22.761053v1
AB - The cyclic nature of many biological systems is frequently reflected in periodic gene expression patterns. Standard dimensionality reductions can reveal such variation but do not link a sample's position on a cycle to an ordering of the molecular features that define it. We introduce clrcycle, which applies a centered log-ratio transformation, learns a cyclic feature ordering by maximizing variance captured by the first Fourier mode, and projects samples onto the corresponding cosine and sine coordinates. The resulting two-dimensional geometry assigns each sample a phase and a radius measuring the magnitude of its projection onto the learned Fourier circle. Across a mouse circadian atlas, clrcycle recovers known temporal organization and reveals tissue-specific differences in rhythmic timing. In FUCCI-profiled single cells, clrcycle represents independently measured cell-cycle phase in two-dimensional coordinates more reliably than conventional two-dimensional embeddings. In human endometrium, a circular coordinate learned from 250 highly variable genes recovers menstrual-stage organization and can stage unseen endometrial biopsies. Together, these results establish clrcycle as powerful, explicitly circular dimensionality reduction technique to quantify high-dimensional cyclic variation across biological cycles.
ER -