Summary: Understanding the “switches” that turn genes on and off is one of the great mysteries of biology. While AI has begun to crack this code, the tools are fragmented and difficult to reuse—until now.
Researchers have revealed CREsted, a powerful software powerhouse. CREsted doesn't just explain how DNA works; it allows scientists to design entirely new, artificial enhancers—short DNA sequences that can program a gene to turn on only in a specific type of cell, such as a neuron or immune cell.
Important Facts
- “All-in-One” Frame: CREsted combines four critical steps—preprocessing, AI model training, interpretation, and synthetic design—to create a single, scalable workflow.
- Cell Type Specification: AI can look at “chromatin accessibility” (what parts of DNA are open and active) to learn exactly what activates an enhancer in one tissue but remains silent in another.
- Proven Versatility: The team has successfully tested CREsted on a variety of systems, including mouse brain, human immune cells, and zebrafish growth.
- Vivo Confirmation: Artificial enhancements designed by AI were physically created and tested on live zebrafish, confirming that AI predictions were accurate in a real biological environment.
- Programmable Biology: By taking off he explained DNA to design in turn, CREsted opens the door to precision medicine, where treatments can be programmed to work only on diseased cells.
Source: VIB
Understanding how genes are turned on and off in specific cell types remains one of the major challenges of biology. Although AI has made great strides in encoding the regulatory logic of DNA, applying these methods across data sets, tissues, and species has remained difficult.
It's new Natural Ways paper, a research team led by Prof. Stein Aerts (VIB & KU Leuven) present CREsted, a software package that enables both the analysis and design of gene regulatory elements in a systematic and step-down manner.
Enhancers are gene regulatory elements—short DNA sequences that control when and where genes are active. Deep learning models can help determine this “control code”, but existing methods are often tied to a single dataset or task, making them difficult to reuse or extend.
To deal with this, Prof. Stein Aerts and his team developed CREsted, a new framework that transforms advanced modeling from a collection of one-off analyzes into a structured and reusable workflow.
“We wanted to go beyond one-off models,” says Niklas Kempynck, a PhD student in Aerts' lab.
“CREsted allows researchers to systematically study advanced thinking across biological systems, from cell-by-cell maps of accessible regulatory DNA and on to sequence design.”
CREsted brings together several steps that are often handled separately: pre-processing, model training, interpretation, and design optimization. It is also designed to fit into existing workflows for single-cell analysis, making it easy for researchers to adopt and implement.
“With CREsted, we offer researchers a complete workflow,” said Dr. Seppe De Winter, who shared the original ownership with Kempynck.
“You can train deep learning models with chromatin accessibility data, interpret what regulatory elements they capture, and then use those models to design new DNA sequences and predicted function for a particular cell type.”
To demonstrate its versatility, the team used CREsted in multiple systems, including mouse brain tissue, human immune cells, cancer cell regions, and developing zebrafish. Across these settings, the framework identified regulatory patterns, predicted enhancer activity, and enabled the design of synthetic enhancers, which were validated in vivo in zebrafish.
To Prof. Stein Aerts, Scientific Director of VIB.AI, CREsted's strength lies in making powerful developments compatible and reusable.
“CRESted makes it very easy to train, interpret, and compare optimization models across data sets,” he says.
“That's important if we want these methods to be widely applicable, not just for understanding regulatory DNA, but also for designing and testing new sequences in a systematic way.”
Taken together, the work shows how AI can help move the field from describing regulatory DNA to actively testing and designing it. With applications ranging from basic biology to biotechnology and medicine, CREsted forms the basis for the systematic and systematic control of gene regulation.
Important Questions Answered:
A: In a way, yes! Just as ChatGPT learns the rules of human language to write a poem, CREsted learns the “control language” of DNA to write an enhancer. It identifies specific patterns (motifs) that tell the cell to “start making this protein now,” allowing scientists to write their own custom genetic instructions.
A: Natural enhancers are often complex and can “leak,” shining light in places it shouldn't. Artificial enhancements designed by CREsted can be “cleaner” and clearer. This is a game changer for gene therapy: imagine a treatment that works only inside the tumor cell while leaving the surrounding healthy tissue completely untouched.
A: That is the main purpose of CREsted. Previously, you had to be an advanced biologist to build these models. CREsted is designed to connect directly to single-cell analytical workflows, making “programmable biology” accessible to broad groups of researchers.
Editor's Notes:
- This article was edited by a Neuroscience News editor.
- The journal paper is fully revised.
- More content has been added by our staff.
About this AI and genetic research news
Author: Gunnar De Winter
Source: VIB
Contact person: Gunnar De Winter – VIB
Image: Image posted in Neuroscience News
Actual research: Open access.
“CREsted: modeling genomic and synthetic cell type-specific enhancers in all tissues and types” by Niklas Kempynck, Seppe De Winter, Casper H. Blaauw, Vasileios Konstantakos, Eren Can Ekşi, Sam Dieltiens, Darina Abaffyová, Valérie Bercier, Valérie Bercier, Gernale Katilskier, Gernale Katilskirie, Ibrahim I. Christiaens, Ludo Van Den Bosch, Lukas Mahieu & Stein Aerts. Natural Ways
DOI:10.1038/s41592-026-03057-2
Abstract
CREsted: modeling genomic and synthetic cell enhancers for specific tissues and species
Sequence-based deep learning models have become the state of the art for analyzing the genomic regulatory code. Especially for developers, these models excel in describing the sequence grammar that is the basis of their work.
To enable end-to-end developer modeling and design, we developed a software package called CREsted (cis-training on the sequence of control elements, definition and design). It includes preprocessing and analysis of a single-cell transposase-accessible chromatin assay using sequencing data, modeling chromatin accessibility from sequencing, sequence architecture and downstream analysis to define promoter grammar.
We demonstrate the performance of CREsted on a mouse cortex and a human blood cell dataset. Additionally, we are using CREsted to compare mesenchymal-like cancer regions between tumor types, and we are investigating strategies to better integrate genomic-based models within CREsted.
Finally, we train the model on the zebrafish enhancer atlas and use this to design and validate cell type-specific enhancers. On a variety of datasets, we show that CREsted facilitates efficient training and analysis, enabling improved concept processing and design of synthetic enhancers across tissues and species.
