Reconstruction of Cellular Trajectories with Biologically Interpretable States
- 1The University of Manchester
Abstract
Elucidating the trajectories of cellular evolution provides biologists with deeper insights into biological systems and disease progression, ultimately facilitating more informed discovery of mechanisms and therapeutic design. Single-cell RNA sequencing (scRNA-seq) is widely used to study cellular evolution. However, it faces two primary limitations. First, because the process destroys cells during gene-expression measurement, it is impossible to track the gene-expression profile of the same cell throughout the entire trajectory. Second, because scRNA-seq captures cells at discrete time points, it provides only static snapshots rather than continuous trajectories, limiting its ability to resolve the full dynamics of cellular transitions. Consequently, various computational approaches have been developed to address these challenges. However, existing models fail to capture the hidden state of each cell, which is essential for reconstructing gene expression at unobserved time points. As the gene expression profiles depend on the cell state, and may vary according to time, we propose a Latent Cell-State Model, where we approximate cell state transition as a Markov chain and obtain the gene expression of each cell at unobserved timepoints depending on the cell state, time, and a latent variable that represents the high-dimensional gene expression profile of a cell. Our model is able to infer biologically meaningful cell states and generate high-quality gene expression profiles at unobserved timepoints, showing potential for accurate trajectory reconstruction.
Related
- FlowMap: Geometry Dynamics Consistent Embedding of RNA Velocity for Interpretable Cellular Trajectories
- From Closed-Set Annotation to In-Context Discovery: Biologically Grounded LLMs for Cellular Profiling
- HiC-LEGO: Biologically Guided High-Resolution 3D Genome Reconstruction Preserves Chromatin Organization at Kilobase Resolution
- Interpretable many-body interaction laws for collective motion from few trajectories
- Multimodal cellular context enables molecular phenotyping from single-molecule trajectories
BibTeX
@misc{wang2026reconstruction,
title = {{Reconstruction of Cellular Trajectories with Biologically Interpretable States}},
author = {Y. Wang and M. Rattray and D. Barber},
howpublished = {bioRxiv},
year = {2026},
month = nov,
doi = {10.64898/2026.11.03.763902},
url = {https://www.biorxiv.org/content/10.64898/2026.11.03.763902v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Reconstruction of Cellular Trajectories with Biologically Interpretable States
AU - Y. Wang
AU - M. Rattray
AU - D. Barber
PY - 2026
DA - 2026/11/03
PB - bioRxiv
DO - 10.64898/2026.11.03.763902
UR - https://www.biorxiv.org/content/10.64898/2026.11.03.763902v1
AB - Elucidating the trajectories of cellular evolution provides biologists with deeper insights into biological systems and disease progression, ultimately facilitating more informed discovery of mechanisms and therapeutic design. Single-cell RNA sequencing (scRNA-seq) is widely used to study cellular evolution. However, it faces two primary limitations. First, because the process destroys cells during gene-expression measurement, it is impossible to track the gene-expression profile of the same cell throughout the entire trajectory. Second, because scRNA-seq captures cells at discrete time points, it provides only static snapshots rather than continuous trajectories, limiting its ability to resolve the full dynamics of cellular transitions. Consequently, various computational approaches have been developed to address these challenges. However, existing models fail to capture the hidden state of each cell, which is essential for reconstructing gene expression at unobserved time points. As the gene expression profiles depend on the cell state, and may vary according to time, we propose a Latent Cell-State Model, where we approximate cell state transition as a Markov chain and obtain the gene expression of each cell at unobserved timepoints depending on the cell state, time, and a latent variable that represents the high-dimensional gene expression profile of a cell. Our model is able to infer biologically meaningful cell states and generate high-quality gene expression profiles at unobserved timepoints, showing potential for accurate trajectory reconstruction.
ER -