Scientists from the Centre for Brain and Disease Research at Leuven University and at VIB.AI have succeeded in creating synthetic DNA using artificial intelligence. The AI model was able to design 'enhancers', or 'switches', that can be used to activate specific genes. "Groundbreaking," the team says. The results of the research have been published in the renowned scientific journal Nature.
All the cells in our body, from our eyes to our stomach, all have the same DNA, but not all this DNA is utilised in every cell type.
Gene activity is determined by a set of instructions within the DNA, called the regulatory code. Here's the thing. It has long been known of DNA that the sequence of base pairs represented by the letters A, C, G and T ensure the production of certain proteins, which are required for the functioning of all human cells.
A piece of DNA that codes for a particular trait forms a gene. Which genes are activated at what time depends on a process called the 'regulatory code'. The logic behind that code has been fodder for research for decades.
"A team led by Professor Stein Aerts and Dr Ibrahim I. Taskiran has now trained a deep learning model to crack this code. This model was able to decipher the enhancer code, providing unprecedented new insights," the researchers reported today.
From the code to the 'switches'
A crucial element in the regulatory code, are the 'enhancers', or 'switches'. Enhancers are pieces of DNA located around genes, on which proteins bind. When that happens, they activate a particular gene. A cell can find expression when well-defined genes are activated, all of which must be precisely regulated by their enhancers.
Professor Stein Aerts' team at VIB-KU Leuven has been working on the regulatory code for more than 20 years. "Step by step, we have gained more insight into the code," says Aerts. "Since 2015, we have been working with deep learning models. This has enormously accelerated the research."
One step further
Because each enhancer consists of several hundred base pairs, the different possibilities of sequence, or order, are virtually endless. By giving the AI model more data and improving it, it was able to predict which enhancer sequence was active in a cell type and which was not.
"Then we went one step further and made adjustments to the enhancers using experiments: for example, we omitted a base or added one," says Professor Aerts. "That way we could quickly find out what impact that had. The AI model thus provides insights into how the synthetic enhancers are constructed and which proteins do or do not bind to them."
The model was tested on fruit fly brain cells and on a type of human skin cell. Because fruit flies are so small and their DNA is easy to manipulate, geneticists often make their first discoveries using this model organism.
"The team then also managed to use their new approach successfully to design enhancers for human DNA".
Eventually, the aim is to read out enhancers for every human cell: in the meantime, the researchers are working on different types of cancer cells and human brain cells.
The findings are interesting for general DNA research but are also crucial for future gene therapies.
"Future therapies will require that certain genes are activated in certain cells but not in others. Thanks to the information from our AI model, the safety of gene therapy can be increased."
Want to know more? Click here for the study in the journal Nature . The work received support from the European Research Council and the Scientific Research Fund Flanders.