Leuven, 2 April 2026 – Unraveling the complex regulatory mechanisms dictating gene activity across diverse cell types has long challenged biologists. Genes are not merely static blueprints but are dynamically turned on or off in specific cellular contexts, driving development, function, and response to environmental signals. At the heart of this precision lie enhancers—short segments of DNA that govern where and when genes are expressed. Despite tremendous advances in sequencing technologies and functional genomics, deciphering the enhancer code remains a formidable task.
In recent years, artificial intelligence, particularly deep learning, has revolutionized our ability to predict and interpret patterns within regulatory DNA. However, existing AI methodologies frequently fall short of generalizing across datasets, tissues, or species, limiting their impact. This fragmentation hampers the broader application of enhancer modeling, often relegated to isolated tasks or bespoke datasets without scalable workflows. Addressing this fundamental bottleneck, Prof. Stein Aerts and his interdisciplinary team at VIB and KU Leuven have introduced CREsted—a transformative software framework designed to unify and elevate enhancer analysis and synthetic design to new heights.
Published today in the prestigious journal Nature Methods, the CREsted platform epitomizes a shift from ad hoc enhancer models towards a comprehensive, reusable pipeline capable of systematic application to diverse biological systems. Its core innovation lies in seamlessly integrating multiple analytical stages—ranging from preprocessing single-cell chromatin accessibility data, through rigorous deep learning model training and interpretation, culminating in the in silico design of synthetic enhancers with tailored activity profiles. This end-to-end approach empowers researchers with unprecedented capability to not only decode the regulatory lexicon but also to actively engineer enhancer sequences.
Enhancers operate by regulating the accessibility of DNA to transcription factors and the transcriptional machinery, thereby fine-tuning gene expression in a cell type- and tissue-specific manner. However, enhancer sequences often exhibit considerable contextual and evolutionary variation, making their prediction and functional validation challenging. Deep learning approaches have demonstrated significant promise in capturing these nonlinear patterns by learning predictive models directly from raw DNA sequences coupled with epigenomic data, such as chromatin accessibility maps gathered at single-cell resolution. Yet, prior efforts have largely focused on narrowly defined datasets or experimental conditions, limiting cross-study comparability and extensibility.
CREsted builds upon this foundation by standardizing data preprocessing to generate high-quality inputs from single-cell ATAC-seq datasets, a widely used assay to identify open chromatin regions emblematic of enhancer activity. Once data is harmonized, the framework trains convolutional neural networks and other deep learning architectures capable of encoding rich representations of regulatory DNA sequences. Importantly, CREsted includes interpretability modules to elucidate which sequence motifs and features the models leverage—providing mechanistic insight that bridges predictive power with biological understanding.
Beyond prediction, the toolkit integrates generative modeling techniques that enable the rational design of synthetic enhancer sequences predicted to function in a cell-type-specific manner. Designed sequences are optimized for maximal activity in target cellular contexts while minimizing off-target effects. This capacity to reverse-engineer enhancer elements from learned models represents a significant leap forward with deep implications for synthetic biology, gene therapy, and functional genomics.
To validate CREsted’s versatility and robustness, the research team applied it to an impressive array of experimental systems. These included mouse brain tissue, providing insight into the neurogenomic regulatory landscape; human immune cell populations, crucial for understanding immune responses and diseases; diverse cancer cell states to dissect oncogenic regulatory rewiring; and zebrafish embryogenesis as a multicellular developmental model. Across these varied contexts, CREsted not only accurately predicted enhancer activity but also guided the design of synthetic enhancers whose function was experimentally confirmed in vivo within zebrafish embryos.
The modular design of CREsted allows it to dovetail with standard single-cell analysis pipelines, lowering the barrier for adoption among researchers who may lack deep computational expertise. This ease of integration encourages widespread application, fostering reproducibility and comparative studies that were previously intractable. According to Niklas Kempynck, a doctoral researcher contributing to the project, “Our goal was to transcend isolated analyses by creating a reusable workflow that can interrogate enhancer logic across biological systems, linking raw chromatin accessibility data all the way to bespoke synthetic sequence design.”
Dr. Seppe De Winter, another leading contributor, emphasizes the comprehensive nature of CREsted: “Researchers can now train intricate deep learning models on their chromatin data, extract biologically meaningful features to understand regulatory code, and ultimately leverage these insights to generate synthetic enhancers predicted to operate in specific cells. This paradigm fosters an iterative experimental-computational cycle that accelerates discovery.”
Prof. Aerts, who also spearheads VIB.AI, highlights the broader significance: “By making enhancer modeling more interpretable and comparable across datasets, CREsted unlocks the potential for systematic manipulation of gene regulation. Such programmable control is critical not only for basic biological research but also for translating findings into therapeutic applications—ranging from precision gene editing to engineered cellular therapies.”
The inception of CREsted marks a milestone in the convergence of computational biology and synthetic genomics. It empowers scientists with a unified platform to navigate the complexity of regulatory DNA, moving beyond mere description to actionable design and functional testing. As the era of single-cell multi-omics matures and datasets proliferate across species and conditions, tools like CREsted will become indispensable for uncovering fundamental principles of gene regulation as well as engineering novel regulatory elements for biotechnological innovation.
Looking forward, the CREsted team envisions expanding the framework to incorporate additional regulatory layers such as histone modifications, DNA methylation, and 3D genome architecture, further enriching model fidelity. Additionally, coupling CREsted with high-throughput synthetic enhancer screening and CRISPR-based perturbation strategies could facilitate rapid, large-scale validation of model predictions and synthetic constructs. Such integration would accelerate the iterative refinement of enhancer models and enable precise manipulation of gene networks to treat disease or engineer complex biological systems.
In a landscape where decoding and redesigning cell-type-specific gene regulation stands at the frontier of biology, CREsted is poised to be a cornerstone resource. By offering a scalable, interpretable, and experimentally validated workflow, it heralds a new chapter in genomic research where artificial intelligence and synthetic biology converge to illuminate and harness the regulatory genome’s full potential.
Subject of Research: Not applicable
Article Title: CREsted: modeling genomic and synthetic cell type-specific enhancers across tissues and species.
News Publication Date: 2-Apr-2026
Web References: https://crested.readthedocs.io/en/stable/
Keywords: Computational biology, Evolutionary biology, Developmental biology, Genetics, Neuroscience
