Kviman, O. | bioRxiv Science

bioRxiv2026

Phylospatial optimal transport for single-cell and spatial transcriptomics data integration

The spatial organization of evolving cell populations in tumors influences how they progress and respond to treatment. Spatially resolved transcriptomics data may be used to examine the locations of these diverse cell states in tumor sections. However, these technologies often lack single-cell resolution and instead capture groups of cells in spatial spots, making it difficult to resolve cancer clones. In contrast, dissociated single-cell transcriptomics data are better suited for obtaining cell lineages, but do not capture the locations of the cells in the tissue. Here, we introduce SpOTr, an optimal transport approach that leverages the concordance between the evolutionary tree and the spatial distribution of cancer cells to integrate single-cell and spatial transcriptomics data from tumors. SpOTr combines a spot deconvolution model with Fused Gromov-Wasserstein optimal transport, in which cells that are close in the tree are drawn to nearby spots. In simulations and in real spatial lineage tracing data, SpOTr recovered the spatial locations of clones better than methods that ignore the tree. In a human colorectal liver metastasis, SpOTr mapped copy number clones from dissociated cells to spatial spots, showing that it successfully maps the evolutionary history of tumors in space.

bioinformatics↗

bioRxiv2025

LN's t-Test: A Principled Approach to t-Testing in Single-Cell RNA Sequencing

Single-cell RNA sequencing (scRNA-seq) has revolutionized the study of cellular heterogeneity, yet differential gene expression (DGE) analysis remains hindered by inconsistencies in log fold change (LFC) estimation. Existing methods, such as those implemented in Scanpy and Seurat, rely on log-transformed count data with a pseudocount, introducing a bias that compromises the reliability of the statistical inference. In this work, we propose LNs t-test, a novel approach to DGE testing that circumvents these biases by employing a log-Normal (LN) distribution-based LFC estimator. Our method jointly estimates the probability of non-zero expression and the mean of positive expression values, enabling an asymptotically unbiased and normally distributed LFC estimator with corresponding confidence intervals. Through extensive simulation studies, we demonstrate that LNs t-test outperforms competing methods by reducing false discovery rates and providing more accurate effect size estimates. Notably, we leverage stochastic ordering theory to explain why conventional t-tests systematically mis-classify non-differentially expressed genes under realistic variance conditions. Our approach offers a theoretically principled and computationally efficient alternative for DGE analysis in scRNA-seq, with implications for improving the reliability and interpretability of single-cell transcriptomics studies. Code that implements the results is available on GitHub: https://github.com/okviman/DE-ZILN.

bioinformatics↗