Understanding Gene Expression-to-Image Learning: Spatial Structure and Information Fidelity
Abstract
Single-cell RNA sequencing (scRNA-seq) enables cell-level analysis of biological heterogeneity, but learning effective representations from high-dimensional, unordered gene expression profiles remains challenging. Gene Expression-to-Image (GE2I) offers an effective learning strategy by arranging gene values on pixel grids for convolutional neural networks, but its underlying mechanisms remain unclear. We study GE2I on full-transcriptome data while keeping the network and training procedure fixed. Our results reveal a trade-off between spatial smoothness and gene-level fidelity. Structured mappings produce smooth, low-frequency patterns that support faster early optimization, whereas assigning multiple genes to the same pixel obscures expression differences and can reduce predictive performance. This trade-off explains why random mappings can compete with structured layouts when many genes are represented. One-to-one gene-to-pixel assignment improves prediction across the tested visual backbones even when image smoothness decreases. We further find that the benefits of pretraining and augmentation depend on the backbone and mapping. These findings provide practical guidance for pixel allocation and the transfer of vision priors to gene expression data.
Related
- Subcellularly Resolved Single-Cell Embedding Learning with Transcriptomic data, Protein Structure and Localization Information
- Physics-driven self-supervised learning for quantitative high-fidelity structured illumination microscopy
- MF-TLMLP: A multi-channel fusion-based multi-fidelity transfer learning method for predicting the cutting performance of conical picks
- LIMBER: A Hybrid RNA Foundation Model for Predicting and Understanding RNA Structure and Function
- Cell-level integration of gene expression and B cell receptor data captures complementary B cell information
BibTeX
@misc{lian2026understanding,
title = {{Understanding Gene Expression-to-Image Learning: Spatial Structure and Information Fidelity}},
author = {Xiaoqing Lian and Pei Liu and Jiahui Tan and Quan Zou and Chen Lin and Xiangxiang Zeng},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.17699},
eprint = {2610.17699},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.17699},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Understanding Gene Expression-to-Image Learning: Spatial Structure and Information Fidelity
AU - Xiaoqing Lian
AU - Pei Liu
AU - Jiahui Tan
AU - Quan Zou
AU - Chen Lin
AU - Xiangxiang Zeng
PY - 2026
DA - 2026/10/13
PB - arXiv
DO - 10.48550/arxiv.2610.17699
UR - https://arxiv.org/abs/2610.17699
AB - Single-cell RNA sequencing (scRNA-seq) enables cell-level analysis of biological heterogeneity, but learning effective representations from high-dimensional, unordered gene expression profiles remains challenging. Gene Expression-to-Image (GE2I) offers an effective learning strategy by arranging gene values on pixel grids for convolutional neural networks, but its underlying mechanisms remain unclear. We study GE2I on full-transcriptome data while keeping the network and training procedure fixed. Our results reveal a trade-off between spatial smoothness and gene-level fidelity. Structured mappings produce smooth, low-frequency patterns that support faster early optimization, whereas assigning multiple genes to the same pixel obscures expression differences and can reduce predictive performance. This trade-off explains why random mappings can compete with structured layouts when many genes are represented. One-to-one gene-to-pixel assignment improves prediction across the tested visual backbones even when image smoothness decreases. We further find that the benefits of pretraining and augmentation depend on the backbone and mapping. These findings provide practical guidance for pixel allocation and the transfer of vision priors to gene expression data.
ER -