In silico generation of gene expression profiles using diffusion models
5 affiliations
- 1IBISC/CNRS-Inria-LISN, University Paris-Saclay
- 2Diagnostic Image Analysis Group, Department of Medical Imaging, Radboud University Medical Center, The Netherlands;
- 3RadboudUMC - Radboud University Medical Center [Nijmegen] (PO Box 9101, 6500 HB Nijmegen, The Netherlands - Netherlands)
- 4TAU, CNRS-INRIA-LISN, University Paris-Saclay
- 5IBISC, University Paris-Saclay (Univ. Evry)
Abstract
BACKGROUND: RNA-seq data is used for precision medicine (e.g., cancer predictions), which benefits from deep learning approaches to analyze complex gene expression data. However, transcriptomics datasets often have few samples compared to deep learning standards. Synthetic data generation is thus being explored to address this data scarcity. So far, only deep generative models such as Variational Autoencoders (VAEs) and Generative Adversarial Networks (GANs) have been used for this aim. Considering the recent success of diffusion models (DM) in image generation, we propose a diffusion-model-based generation pipeline that leverages the power of such generative models on transcriptomics data. RESULTS: This paper presents two state-of-the-art diffusion models (DDPM and DDIM) and achieves their adaptation in the transcriptomics field. DM-generated data of L1000 landmark genes show better predictive performance over TCGA and GTEx datasets. We also compare linear and nonlinear reconstruction methods to recover the complete transcriptome. Results show that such reconstruction methods can boost the performance of diffusion models, as well as VAEs and GANs. CONCLUSIONS: Overall, the extensive comparison of various generative models using data quality indicators shows that diffusion models rank among the best-performing methods, making them promising synthetic transcriptomics generators.
Related
- Genome wide identification and in silico mRNA expression profiling of the WRKY transcription factor gene family in Salvia hispanica
- Inferring gene expression profiles of tumor clones
- Topology-adjusted knockout (TAKO): interpretable in silico single-gene knockout ranking on wild-type single-cell gene regulatory networks.
- Free-Energy and Position-Dependent Diffusion Profiles Shape Passive Permeation of a Semisynthetic Naphthoquinone across a Model Lipid Bilayer.
- Rice Expression Database (RED) 2.0: an updated rice database integrating diverse RNA expression profiles based on the telomere-to-telomere reference genome
BibTeX
@article{lacan2026silico,
title = {{In silico generation of gene expression profiles using diffusion models}},
author = {Alice Lacan and Romain André and Michèle Sébag and Blaise Hanczar},
journal = {BMC Bioinformatics},
year = {2026},
month = may,
doi = {10.1186/s12859-026-06470-8},
url = {https://doi.org/10.1186/s12859-026-06470-8},
note = {CC BY},
}
RIS
TY - JOUR
TI - In silico generation of gene expression profiles using diffusion models
AU - Alice Lacan
AU - Romain André
AU - Michèle Sébag
AU - Blaise Hanczar
PY - 2026
DA - 2026/05/31
JO - BMC Bioinformatics
DO - 10.1186/s12859-026-06470-8
UR - https://doi.org/10.1186/s12859-026-06470-8
AB - BACKGROUND: RNA-seq data is used for precision medicine (e.g., cancer predictions), which benefits from deep learning approaches to analyze complex gene expression data. However, transcriptomics datasets often have few samples compared to deep learning standards. Synthetic data generation is thus being explored to address this data scarcity. So far, only deep generative models such as Variational Autoencoders (VAEs) and Generative Adversarial Networks (GANs) have been used for this aim. Considering the recent success of diffusion models (DM) in image generation, we propose a diffusion-model-based generation pipeline that leverages the power of such generative models on transcriptomics data. RESULTS: This paper presents two state-of-the-art diffusion models (DDPM and DDIM) and achieves their adaptation in the transcriptomics field. DM-generated data of L1000 landmark genes show better predictive performance over TCGA and GTEx datasets. We also compare linear and nonlinear reconstruction methods to recover the complete transcriptome. Results show that such reconstruction methods can boost the performance of diffusion models, as well as VAEs and GANs. CONCLUSIONS: Overall, the extensive comparison of various generative models using data quality indicators shows that diffusion models rank among the best-performing methods, making them promising synthetic transcriptomics generators.
SN - 1471-2105
ER -