scMaize: A Single-Cell Foundation Model and Integrated Atlas for Maize
Abstract
Single-cell transcriptomics has resolved cell-type-specific gene expression in plants, yet maize still lacks an integrated reference and species-specific foundation models. We present scMaize, combining scMaizeAtlas, an integrated atlas of 385,675 cells from 20 projects and 66 samples across seven tissues with hierarchical annotation, with two Transformer-based foundation models pretrained on this atlas. scMaizeExp serves as an expression-only baseline, while scMaizeGO incorporates Gene Ontology (GO) functional embeddings as an inductive bias. Although global expression-prediction accuracy was comparable, the GO prior improved rank-order prediction, strengthened attention toward functionally coherent gene modules, and enhanced embedding topology, with scMaizeGO achieving 86.0% cell-type and 97.1% tissue classification accuracy. Zero-shot evaluation demonstrated the cross-species generalizability of scMaizeGO representations, and few-shot fine-tuning enabled accurate cross-species classification with minimal labeled data. Root perturbation-condition analysis showed that the model encoded treatment-specific cellular states beyond cell-type identity, with the GO prior amplifying perturbation signals approximately threefold. Expression projection identified condition-responsive genes enriched for known stress pathways, and attention analysis revealed predominantly condition-specific changes in gene-gene attention that were weakly associated with expression-projection changes. An online platform (https://www.scmaize.com) provides atlas exploration, model access, and zero-code analysis tools. scMaize establishes a framework demonstrating that species-specific pretraining with functional priors enables transferable, perturbation-aware representations for crop single-cell genomics.
Related
- An Integrated Single-Nucleus Atlas Resolves Cell-Type-Specific Programs and Molecular Subtypes in Alzheimer's Disease
- LiverHomo: An Integrated and Interactive Human Liver Single-Cell Atlas with Protein Interaction and Cell-Cell Communication Networks
- AREION: A Foundation Model for Atlas-Agnostic Representation Learning of Brain Connectome
- NeoOmni Atlas: an integrated database for pan-cancer neoantigen derived from proteogenomics analysis
- Integrated Transcriptome and MicroRNAome Analysis Reveals Dynamic Responses Underlying ZmWAK-Mediated Resistance to Head Smut in Maize
BibTeX
@misc{cheng2026scmaize,
title = {{scMaize: A Single-Cell Foundation Model and Integrated Atlas for Maize}},
author = {Q. Cheng and Y. Zhang and H. T. Wu and A. Zhao and Q. M. Shang and F. X. Wang and J. Yan},
howpublished = {bioRxiv},
year = {2026},
month = aug,
doi = {10.64898/2026.08.01.742180},
url = {https://www.biorxiv.org/content/10.64898/2026.08.01.742180v2},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - scMaize: A Single-Cell Foundation Model and Integrated Atlas for Maize
AU - Q. Cheng
AU - Y. Zhang
AU - H. T. Wu
AU - A. Zhao
AU - Q. M. Shang
AU - F. X. Wang
AU - J. Yan
PY - 2026
DA - 2026/08/01
PB - bioRxiv
DO - 10.64898/2026.08.01.742180
UR - https://www.biorxiv.org/content/10.64898/2026.08.01.742180v2
AB - Single-cell transcriptomics has resolved cell-type-specific gene expression in plants, yet maize still lacks an integrated reference and species-specific foundation models. We present scMaize, combining scMaizeAtlas, an integrated atlas of 385,675 cells from 20 projects and 66 samples across seven tissues with hierarchical annotation, with two Transformer-based foundation models pretrained on this atlas. scMaizeExp serves as an expression-only baseline, while scMaizeGO incorporates Gene Ontology (GO) functional embeddings as an inductive bias. Although global expression-prediction accuracy was comparable, the GO prior improved rank-order prediction, strengthened attention toward functionally coherent gene modules, and enhanced embedding topology, with scMaizeGO achieving 86.0% cell-type and 97.1% tissue classification accuracy. Zero-shot evaluation demonstrated the cross-species generalizability of scMaizeGO representations, and few-shot fine-tuning enabled accurate cross-species classification with minimal labeled data. Root perturbation-condition analysis showed that the model encoded treatment-specific cellular states beyond cell-type identity, with the GO prior amplifying perturbation signals approximately threefold. Expression projection identified condition-responsive genes enriched for known stress pathways, and attention analysis revealed predominantly condition-specific changes in gene-gene attention that were weakly associated with expression-projection changes. An online platform (https://www.scmaize.com) provides atlas exploration, model access, and zero-code analysis tools. scMaize establishes a framework demonstrating that species-specific pretraining with functional priors enables transferable, perturbation-aware representations for crop single-cell genomics.
ER -