AI-designed synthetic DNA precisely controls gene expression

Artificial intelligence (AI) methods have produced simulated DNA to determine gene expression, a new study has revealed.

Cis-regulatory elements (CREs) are segments of the DNA genome which do not code for genes but instead control how nearby genes are expressed. They feature binding sites for transcription factors enabling specific genes to generate certain proteins. Compared to genes, there is a larger variety of DNA sequences that code for CREs, making it harder to decipher and determine the particular function of a CRE and the genes that it affects.

'This project essentially asks the question: Can we learn to read and write the code of these regulatory elements? If we think about it in terms of language, the grammar and syntax of these elements is poorly understood. And so, we tried to build machine learning methods that could learn a more complex code than we could do on our own,' said Dr Steven Reilly, one of study's senior authors and assistant professor of genetics at Yale School of Medicine, Connecticut.

Gene therapy aims to treat illnesses by modifying a person's DNA by adding missing genes, replacing defective ones, or silencing harmful genes. Since every cell in the human body contains a person's entire DNA, initially proposed gene therapies courted concerns over the potentially severe side effects if the therapies affected non-targeted tissues.

Researchers from the Yale School of Medicine, the Jackson Laboratory, Maine, and the Broad Institute of Massachusetts Institute of Technology and Harvard University created an AI model called Computational Optimisation of DNA Activity (CODA). The model was developed using 'deep learning', which employs a structure similar to neural networks, to understand the DNA of existing CREs and then designed 775,000 synthetic CREs to express genes exclusively in the brain, liver or blood cells.

Published in Nature, the research trialled the effectiveness of some of their synthetic CREs in mice and zebrafish models. One particular synthetic CRE successfully activated a specific layer of cells in a mouse's brain but not elsewhere, even though the CRE was delivered to all the cells in the mouse's body. Another CRE selectively activated a fluorescent protein within a zebrafish's liver.

The researchers were sceptical that the synthetic CREs would be as effective as intended, with many expecting severe adverse effects from altered gene expression across different bodily tissues. However, the synthetic CREs indicated a surprising higher level of selectivity, surpassing that of many naturally occurring CREs for designated cell types.

A future aim of the study is to cultivate CREs beyond those naturally found in human DNA, whose function – some researchers believe – has been curtailed by evolution.

'There are a lot of potential solutions out there for lots of different possible things you might want a regulatory element to do. Evolution maybe has never wanted to build a really great driver for an Alzheimer's drug, but that doesn't mean it can't exist,' said Dr Reilly.