arxiv_cv 90% Match Research Paper Computational Biologists,Bioinformaticians,Medical Researchers,AI Researchers (Medical Imaging) 3 days ago

GeneFlow: Translation of Single-cell Gene Expression to Histopathological Images via Rectified Flow

computer-vision › medical-imaging

📄 Abstract

Abstract: Spatial transcriptomics (ST) technologies can be used to align transcriptomes with histopathological morphology, presenting exciting new opportunities for biomolecular discovery. Using ST data, we construct a novel framework, GeneFlow, to map transcriptomics onto paired cellular images. By combining an attention-based RNA encoder with a conditional UNet guided by rectified flow, we generate high-resolution images with different staining methods (e.g. H&E, DAPI) to highlight various cellular/tissue structures. Rectified flow with high-order ODE solvers creates a continuous, bijective mapping between transcriptomics and image manifolds, addressing the many-to-one relationship inherent in this problem. Our method enables the generation of realistic cellular morphology features and spatially resolved intercellular interactions from observational gene expression profiles, provides potential to incorporate genetic/chemical perturbations, and enables disease diagnosis by revealing dysregulated patterns in imaging phenotypes. Our rectified flow-based method outperforms diffusion-based baseline method in all experiments. Code can be found at https://github.com/wangmengbo/GeneFlow.

Authors (9)

Mengbo Wang

Shourya Verma

Aditya Malusare

Luopin Wang

Yiyang Lu

Vaneet Aggarwal

+3 more

Submitted

October 31, 2025

arXiv Category

q-bio.QM

arXiv PDF

Key Contributions

Introduces GeneFlow, a novel framework that maps spatial transcriptomics data to paired cellular images using an attention-based RNA encoder and a conditional UNet guided by rectified flow. This method generates high-resolution images with different staining, enabling visualization of cellular structures and intercellular interactions, and addressing the many-to-one mapping problem.

Business Value

Accelerates biomedical discovery by enabling researchers to visualize cellular and tissue structures directly from gene expression data, potentially leading to faster drug development, improved diagnostics, and a deeper understanding of biological processes.

Paper Metadata

Innovation Type

Algorithmic Framework

Deployment Feasibility

Moderate. Requires specialized biological data (ST) and significant computational resources for training and inference.

Limitations Addressed

The challenge of aligning transcriptomes with histopathological morphology, the many-to-one relationship between gene expression and image features, and the need for high-resolution visualization of cellular structures and interactions.

Technical Tags

Spatial TranscriptomicsHistopathologyImage GenerationRectified FlowUNetRNA EncoderODE SolversBiomolecular Discovery

Research Topics

Medical ImagingBioinformaticsGenerative ModelsComputational BiologyImage Synthesis

Methods & Architectures

GeneFlow frameworkAttention-based RNA EncoderConditional UNetRectified FlowHigh-order ODE Solvers UNetEncoder-Decoder

Applications & Tasks

Biomedical Research Drug Discovery Disease Diagnosis Genomics Mapping Gene Expression to MorphologyGenerating Histopathological ImagesUnderstanding Cellular Interactions Generating high-resolution histopathological images from ST dataVisualizing cellular/tissue structuresEnabling disease diagnosis through morphology prediction

Related Fields

Computational BiologyMedical ImagingBioinformaticsMachine LearningGenerative Models

Keywords

Spatial TranscriptomicsHistopathologyImage GenerationRectified FlowUNetRNAGene ExpressionCellular MorphologyBiomedicalDisease DiagnosisODEGenerative Models

Academic Context

#Medical Imaging#Bioinformatics#Generative Models#Computational Biology#Image Synthesis

Commercial Potential

Potential Products

Diagnostic Tools for CancerDrug Discovery PlatformsResearch Visualization Software

Target Industries

BiotechnologyPharmaceuticalsHealthcareResearch Institutions

Use Case Examples

Visualizing tumor microenvironments from gene expression dataPredicting cellular responses to drug treatmentsGenerating synthetic histopathology images for training diagnostic models

Competitive Edge

Offers a novel approach to bridging the gap between molecular data (gene expression) and visual data (histopathology) using advanced generative techniques like rectified flow, providing a more direct and continuous mapping than previous methods.

Market Opportunity

Significant, driven by the growing fields of precision medicine and AI in healthcare.

Revenue Models

Licensing of the technology to biotech/pharma companiesdevelopment of diagnostic software.

Resource Requirements

Compute Needs

High, especially for training deep learning models and ODE solvers.

Data Requirements

Spatial transcriptomics data paired with corresponding cellular images.

Deployment Constraints

Availability and quality of paired ST and image data.

Scalability

The use of ODE solvers and UNet suggests potential for scalability, but training complexity is high.

Regulatory Considerations

Potential for use in medical diagnostics requires regulatory approval (e.g.FDA).

Production Readiness

Maturity Level

Research

Time to Market

3-5 years for clinical applications.

Patent Potential

Moderate, particularly for the GeneFlow framework and its specific components.

View Full Paper Back to Papers