In a significant leap forward for genomics, MIT chemists have developed a technique to swiftly predict the 3D structure of genomes, using generative artificial intelligence. This new method, capable of calculating countless structures within mere minutes, surpasses the slower, traditional approaches that require about a week to analyze the structure from a single cell. MIT News reports that this innovation could pave the way to deeply understand how gene expression in individual cells is influenced by the genome's spatial organization.

The AI model, known as ChromoGen, is composed of a deep learning system that "reads" genetic sequences and a generative AI trained with over 11 million chromatin conformations. Its ability to quickly analyze long DNA strands has not only cut down the time to predict possible structures but is set to radically change the future of genomics research. "Our goal was to try to predict the three-dimensional genome structure from the underlying DNA sequence," Bin Zhang, an MIT associate professor of chemistry and senior author of the study, told MIT News. "Now that we can do that, which puts this technique on par with the cutting-edge experimental techniques, it can really open up a lot of interesting opportunities."

Typically, researchers use a technique called Hi-C, a labor-intensive process for mapping out the chromatin structure in the cell nucleus, where DNA is packed tightly. The ChromoGen model accomplishes a similar task in a fraction of the time. As Greg Schuette, an MIT graduate student and lead author of the study, explains, "Whereas you might spend six months running experiments to get a few dozen structures in a given cell type, you can generate a thousand structures in a particular region with our model in 20 minutes on just one GPU." Such an increase in efficiency could prove instrumental in accelerating the pace of genetic research, as per MIT News.

Upon completion of the model's training phase, the team was able to generate structure predictions for over 2,000 DNA sequences and found that their model's outputs were either identical or highly similar to those obtained through experimental data. This milestone indicates that the model could be potentially applicable across a wide variety of cell types and mutations, which can drastically impact understanding of chromatin conformation changes linked to disease. "There are a lot of interesting questions that I think we can address with this type of model," Zhang further shared with MIT News.

The advance holds promise for bridging the gap between physical experiments and computational predictions in the field, offering a fast and accurate method for visualizing how genes are spatially organized in cells and potentially leading to new approaches for the study of diseases at the genomic level. The researchers have made their data and the ChromoGen model available to the wider scientific community, a move that is likely to foster further research and development in this area. The study was funded by the National Institutes of Health, highlighting its significance in advancing our understanding of genetic material and its influence on cell behavior.