Summary: Understanding the molecular “switches” that control when and where genes turn on remains a major challenge in biology. Recent advances in artificial intelligence are beginning to decode this regulatory code, but tools have often been fragmented and hard to reuse. A new software suite, CREsted, unifies these steps into a single, scalable framework that both decodes enhancer logic and designs synthetic enhancers with cell-type-specific activity.
CREsted is a comprehensive toolkit for modeling and engineering enhancers—short DNA elements that direct gene activation in specific cell types. Rather than offering a single-purpose solution, CREsted combines data preprocessing, deep learning model training, model interpretation, and sequence design into an integrated pipeline that fits into common single-cell analysis workflows. This makes it easier for researchers to move from raw chromatin accessibility maps to interpretable models and, ultimately, to synthetically designed sequences predicted to be active in chosen cell types.
Key Facts
- All-in-one framework: CREsted consolidates preprocessing, model training, feature interpretation, and synthetic sequence design into one reproducible workflow.
- Cell-type specificity: The models learn from chromatin accessibility maps which DNA sequence features distinguish enhancers that are active in one cell type but inactive in others.
- Cross-system utility: The framework was evaluated on diverse datasets, including mouse cortex, human peripheral blood immune cells, mesenchymal-like cancer cell states, and zebrafish development.
- In vivo validation: Synthetic enhancers designed with CREsted were built and tested in living zebrafish, confirming the models’ predictions in a real biological context.
- Path to programmable biology: By enabling sequence design rather than purely descriptive modeling, CREsted supports applications in basic research, biotechnology, and eventually precision therapies that activate only in targeted cell types.
Source: VIB
Turning the regulatory rules encoded in DNA into actionable designs is a major step forward. Deep learning has advanced our ability to read sequence grammar underlying enhancer function, but prior methods were often one-off analyses tied to specific datasets or tasks. CREsted addresses this fragmentation by providing a reusable, extensible software package for end-to-end enhancer modeling and design.
Developed by a team led by Prof. Stein Aerts (VIB & KU Leuven), CREsted—short for cis-regulatory element sequence training, explanation and design—accepts single-cell ATAC-seq (chromatin accessibility) data, trains sequence-based models to predict accessibility and enhancer activity, interprets which motifs and features drive predictions, and uses those models to generate synthetic sequences with intended cell-type-specific activity.
As the developers explain, the goal was to move beyond isolated, bespoke models toward a standardized workflow that researchers can apply across tissues and species. CREsted is designed to plug into existing single-cell analysis pipelines so experimentalists and computational researchers can adopt it with less overhead.
The team demonstrated CREsted’s versatility by applying it to several biological contexts. Models trained on mouse cortex and human peripheral blood mononuclear cells revealed regulatory patterns and predictive sequence features. The framework was used to compare mesenchymal-like cancer cell states across tumor types and to explore fine-tuning strategies for genomic foundation models. Finally, a model trained on a zebrafish developmental atlas was used to design synthetic enhancers that were validated in vivo.
For labs interested in programmable control of gene expression, CREsted offers a practical route from data to design. It enables systematic comparisons of enhancer models across datasets, clarifies the sequence grammar underlying cell-type-specific activity, and produces candidate synthetic sequences that can be tested experimentally.
Key Questions Answered:
A: In a conceptual sense, yes. Like language models that learn grammar and style to generate text, CREsted learns regulatory sequence patterns that define enhancer activity and can generate new sequences predicted to drive expression in specific cell types.
A: Natural enhancers are often complex and can show unintended activity outside their target cells. Synthetic enhancers can be engineered to be more specific and less “leaky,” which is valuable for targeted gene therapies or precise experimental control.
A: Accessibility is a core aim. CREsted is built to integrate with standard single-cell workflows, lowering the barrier so more research groups—not only specialists in computational genomics—can train models, interpret results, and design sequences.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- The cited journal paper was reviewed in full.
- Additional context was added by the editorial staff.
About this AI and genetics research news
Author: Gunnar De Winter
Source: VIB
Contact: Gunnar De Winter – VIB
Image: The image is credited to Neuroscience News
Original Research: Open access.
“CREsted: modeling genomic and synthetic cell type-specific enhancers across tissues and species” by Niklas Kempynck, Seppe De Winter, Casper H. Blaauw, Vasileios Konstantakos, Eren Can Ekşi, Sam Dieltiens, Darina Abaffyová, Valérie Bercier, Ibrahim I. Taskiran, Gert Hulselmans, Katina Spanier, Valerie Christiaens, Ludo Van Den Bosch, Lukas Mahieu & Stein Aerts. Nature Methods
DOI: 10.1038/s41592-026-03057-2
Abstract
CREsted: modeling genomic and synthetic cell type-specific enhancers across tissues and species
Sequence-based deep learning models are now the leading approach for decoding the genomic regulatory code, particularly for enhancer elements where they can resolve the sequence grammar behind activity. To support end-to-end enhancer analysis and design, the authors developed CREsted, a software package that integrates preprocessing and analysis of single-cell ATAC-seq data, sequence-based modeling of chromatin accessibility, sequence design, and downstream interpretation to uncover enhancer grammar.
The paper demonstrates CREsted on multiple datasets, including mouse cortex and human peripheral blood mononuclear cells, compares mesenchymal-like cancer cell states across tumor types, evaluates fine-tuning approaches for genomic foundation models, and trains a model on a zebrafish development atlas to design and validate cell-type-specific enhancers in vivo. Across these applications, CREsted streamlines model training and interpretation, enabling systematic investigation of enhancer logic and the design of synthetic enhancers across tissues and species.