spaTransfer: transfer learning for single-cell and spatial transcriptomics data using non-negative matrix factorization
Abstract
Recent advances in spatially-resolved transcriptomics have enabled profiling of gene expression in a spatial context, which has led to the generation of large-scale single-cell and spatial atlases with computationally-derived cell type or spatial domain labels. An increasingly important task with these data has become the transfer of cell type or spatial domain annotations from a given reference (or source) atlas into a new target tissue or sample. The reference and target datasets could be at different resolutions or measured on different experimental platforms. Here, we present a method to perform cross-platform transfer learning that takes as input single-cell or spatial domain labels from a reference atlas or dataset and transfers the labels to a target dataset at a similar or different resolution. Specifically, we use non-negative matrix factorization (NMF) on the reference data to identify factors associated with labels of interest and project these factors into the target dataset to label each new observation. We use a multinomial model with the factors as covariates and labels as the response to predict labels in the target dataset. In contrast to existing approaches, the advantage of our approach is interpretability, without compromising on accuracy. We demonstrate the performance of the method in two human brain tissues and show that our model identifies spatially coherent domains in the target datasets with concordance of marker gene expression. We implement spaTransfer in open-source software as an R package (github.com/cindyfang70/spaTransfer).
Related
- Integrative Transcriptomics and Machine Learning Nominate IL15- and PPP2R1A-Centered Programs in PTSD Using Pathway-Informed Autonomic–Cardiac Gene Prioritization
- Geometric-aware deep learning for deciphering tissue structure from spatially resolved transcriptomics
- BRIDGE-KO: auditable transfer learning with bridged knockoffs improves cross-cohort microbiome prediction of immunotherapy response
- Automated phytoplankton single‐cell counting from digital images with transfer learning for harmful algal bloom monitoring in the English Channel and North Sea
- Machine-learning-assisted integration of single-cell and bulk transcriptomics characterizes glioblastoma ecosystem states and immune-associated programs
BibTeX
@misc{fang2026spatransfer,
title = {{spaTransfer: transfer learning for single-cell and spatial transcriptomics data using non-negative matrix factorization}},
author = {Cindy Fang and Kelsey D. Montgomery and Sarah E. Maguire and Anthony D. Ramnauth and Boyi Guo and Ryan Miller and Joel E. Kleinman and Thomas M. Hyde and Keri Martinowich and Kristen R. Maynard and Stephanie C. Page and Stephanie C. Hicks},
howpublished = {bioRxiv},
year = {2026},
month = oct,
doi = {10.64898/2025.12.12.694021},
url = {https://doi.org/10.64898/2025.12.12.694021},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - spaTransfer: transfer learning for single-cell and spatial transcriptomics data using non-negative matrix factorization
AU - Cindy Fang
AU - Kelsey D. Montgomery
AU - Sarah E. Maguire
AU - Anthony D. Ramnauth
AU - Boyi Guo
AU - Ryan Miller
AU - Joel E. Kleinman
AU - Thomas M. Hyde
AU - Keri Martinowich
AU - Kristen R. Maynard
AU - Stephanie C. Page
AU - Stephanie C. Hicks
PY - 2026
DA - 2026/10/24
PB - bioRxiv
DO - 10.64898/2025.12.12.694021
UR - https://doi.org/10.64898/2025.12.12.694021
AB - Recent advances in spatially-resolved transcriptomics have enabled profiling of gene expression in a spatial context, which has led to the generation of large-scale single-cell and spatial atlases with computationally-derived cell type or spatial domain labels. An increasingly important task with these data has become the transfer of cell type or spatial domain annotations from a given reference (or source) atlas into a new target tissue or sample. The reference and target datasets could be at different resolutions or measured on different experimental platforms. Here, we present a method to perform cross-platform transfer learning that takes as input single-cell or spatial domain labels from a reference atlas or dataset and transfers the labels to a target dataset at a similar or different resolution. Specifically, we use non-negative matrix factorization (NMF) on the reference data to identify factors associated with labels of interest and project these factors into the target dataset to label each new observation. We use a multinomial model with the factors as covariates and labels as the response to predict labels in the target dataset. In contrast to existing approaches, the advantage of our approach is interpretability, without compromising on accuracy. We demonstrate the performance of the method in two human brain tissues and show that our model identifies spatially coherent domains in the target datasets with concordance of marker gene expression. We implement spaTransfer in open-source software as an R package (github.com/cindyfang70/spaTransfer).
SN - 2692-8205
ER -