Gene regulation is far more predictable than previously believed, scientists conclude after developing deep learning model PARM. This might bring an end to a scientific mystery: how genes know when to switch on or off. Today, scientists publish in Nature about their relentless back-and-forth between lab experiments and computation that enabled them to build this lightweight model. Scientists around the world can now start using this tool for reading these genetic instructions, creating leads for new cancer diagnostics, patient stratification, and future therapies.
“The classical genetic code explains how genes in our DNA encode proteins,” explains Bas van Steensel , group leader at the Netherlands Cancer Institute (NKI) and Oncode Institute and co-corresponding author on the paper. “But for most genes, we honestly didn’t understand how they are regulated. We know that the DNA between our genes contains regulatory elements such as promotors. However, the language of this control system that decides whether a gene turns on or off, in which cell, and how strongly was largely unknown.”
Bold mission
At the same time, most cancer related mutations are located in the non-coding part of our genome, illustrating the immense relevance of this unsolved issue. Until now, interpreting such mutations has been extremely difficult. With the PARM model this becomes possible.
Starting on a bold mission to decode the genomes operating system, seven research groups joined forces in Oncode Institute’s PERICODE project. A technology developed in the Bas van Steensel lab at the NKI enabled measuring gene regulation at an unprecedented scale. Millions of carefully controlled measurements captured how short DNA sequences influence gene activity.
Super-efficient
But data alone is not insight. That is where Jeroen de Ridder’s research group from UMC Utrecht and Oncode Institute entered the picture. The volume of data specifically targeted to gene regulation enabled training AI models that truly captured the biological rules underlying gene activation. “Most AI models learn from whatever data happens to exist,” de Ridder explains. “Here, the measurements and the AI were designed together. This allowed us to make super-efficient models for specific cell types that could be applied at a scale previously unthinkable”
Rigorous testing
The new model enabled the team to predict how gene regulation differs between cell types and how it changes when cells are exposed to stimuli such as specific drugs. Moreover, the model revealed in extreme detail what the architecture of the ‘on and off buttons’ of each gene is. Crucially, the team did not stop at prediction. Every model output was subjected to rigorous experimental testing to make sure that these predictions were indeed correct.
“We can now actually read the language of the gene control system”, says Van Steensel. “Our PARM model allows us to uncover these rules at scale, so we can now understand, and even predict, how regulatory DNA controls gene activity.”
Lightweight
Despite notable progress in the field, the existing AI models were either too heavy to be applied to the vast numbers of mutations that exist or are too generic and do not adequately capture cell type variability. The PARM model changes that. It allows researchers to predict the functional impact of regulatory mutations in specific cell types and under specific conditions, such as drug treatments, opening new paths for cancer diagnostics, patient stratification, and future therapies.
Last week, Google’s Deepmind published in Nature about their model AlphaGenome, aimed at understanding gene regulation as well. “This is a great model”, says Van Steensel. “However, PARM is more flexible and it is experimentally and computationally lightweight. The tool requires around 1000 times less computing power than AlphaGenome, making it far more feasible for academic researchers around the world. With this model you only need one petridish of cells and one day of computing to see in detail how a particular cell type, such as a tumor cell, uses its DNA code to respond to a signal such as a hormone, nutrient or drug.”
Joining forces
The PARM model was developed within the PERICODE project, initiated by Oncode Institute . Seven research groups collaborated within the project: Bas van Steensel (NKI), Jeroen de Ridder (UMCU), Emile Voest (NKI), Michiel Vermeulen (NKI), Lude Franke (UMCG), Sarah Derks (Amsterdam UMC), Wilbert Zwart (NKI). The AVL Foundation financially supported the PERICODE project.
The Netherlands Cancer Institute
The Netherlands Cancer Institute (NKI) is among the world’s foremost comprehensive cancer centers, combining innovative fundamental, translational, and clinical research with dedicated patient care. In our research institute, around 750 researchers from 45 countries work towards solving the mysteries of health and disease and improving the prospects of cancer patients. We gratefully acknowledge funding from the Dutch Ministry of Health, Welfare and Sport, the Dutch Cancer Society, and individual donors.
Nature
Experimental study
Cells
Regulatory grammar in human promoters uncovered by MPRA-based deep learning
3-Feb-2026