LOTUS: lineage-aware optimal transport reconstructs continuous clonal fate dynamics
- 1Center for Quantitative Biology, Peking University, Beijing, 100871, China. Center for Machine Learning Research, Peking University. AI for Science Institute, B
Abstract
Destructive sampling in single-cell sequencing obscures ancestor-descendant relationships, limiting the continuous reconstruction of cell-fate dynamics. Here we present Lineage-aware optimal transport for unbalanced single-cell dynamics (LOTUS), a framework that integrates time-resolved transcriptomic snapshots with clonal lineage information from engineered barcodes or endogenous T cell receptor clonotypes. LOTUS couples clone-level optimal transport with neural unbalanced transport to model continuous cellular state change, population growth and clonal propagation under lineage-consistent constraints. In branching simulations, LOTUS reduced lineage-inconsistent transport relative to benchmark methods, lowering off-target routing from 4.94-14.84% to 1.10%. On LARRY lineage-barcode scRNA-seq of hematopoiesis, LOTUS reconstructed branching trajectories, improved early neutrophil-monocyte fate-boundary prediction, and inferred multi-fate progenitors and fate-biased meta-clones, linking their distinct fate potentials to inferred epigenetic priming. On CellTagging scRNA-seq of fibroblast reprogramming, it distinguished reprogrammed from failed trajectories and prioritized fate-associated perturbation genes. In longitudinal scRNA-seq and TCR-seq of malaria-responsive memory (CD45RA-) CD4+ T cells, it preserved endogenous clonotype structure and reconstructed TR1 dynamics and regulatory networks. LOTUS thus converts lineage-resolved single-cell snapshots into interpretable, continuous models of fate dynamics and their regulatory programs.
Related
- Molecular Characterization of T-Lineage Acute Lymphoblastic Leukemia by an Optimal-Transport Based Multi-Omics Integration Framework
- CloneOT: predicting clone-specific perturbation responses with lineage-supervised neural inverse optimal transport
- scMTG reconstructs single-cell temporal dynamics with Markov transition generators
- Learning developmental dynamics from lineage context
- Locus-specific spatial and transcriptional dynamics direct T cell lineage choice in the thymus
BibTeX
@misc{lu2026lotus,
title = {{LOTUS: lineage-aware optimal transport reconstructs continuous clonal fate dynamics}},
author = {J. Lu and H. Hou and Z. Zhang and Y. Lin and Y. Lin and P. Zhou},
howpublished = {bioRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.28.762359},
url = {https://www.biorxiv.org/content/10.64898/2026.10.28.762359v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - LOTUS: lineage-aware optimal transport reconstructs continuous clonal fate dynamics
AU - J. Lu
AU - H. Hou
AU - Z. Zhang
AU - Y. Lin
AU - Y. Lin
AU - P. Zhou
PY - 2026
DA - 2026/10/28
PB - bioRxiv
DO - 10.64898/2026.10.28.762359
UR - https://www.biorxiv.org/content/10.64898/2026.10.28.762359v1
AB - Destructive sampling in single-cell sequencing obscures ancestor-descendant relationships, limiting the continuous reconstruction of cell-fate dynamics. Here we present Lineage-aware optimal transport for unbalanced single-cell dynamics (LOTUS), a framework that integrates time-resolved transcriptomic snapshots with clonal lineage information from engineered barcodes or endogenous T cell receptor clonotypes. LOTUS couples clone-level optimal transport with neural unbalanced transport to model continuous cellular state change, population growth and clonal propagation under lineage-consistent constraints. In branching simulations, LOTUS reduced lineage-inconsistent transport relative to benchmark methods, lowering off-target routing from 4.94-14.84% to 1.10%. On LARRY lineage-barcode scRNA-seq of hematopoiesis, LOTUS reconstructed branching trajectories, improved early neutrophil-monocyte fate-boundary prediction, and inferred multi-fate progenitors and fate-biased meta-clones, linking their distinct fate potentials to inferred epigenetic priming. On CellTagging scRNA-seq of fibroblast reprogramming, it distinguished reprogrammed from failed trajectories and prioritized fate-associated perturbation genes. In longitudinal scRNA-seq and TCR-seq of malaria-responsive memory (CD45RA-) CD4+ T cells, it preserved endogenous clonotype structure and reconstructed TR1 dynamics and regulatory networks. LOTUS thus converts lineage-resolved single-cell snapshots into interpretable, continuous models of fate dynamics and their regulatory programs.
ER -