Dr. Itai Yanai is Using the Data-Crunching Power of Machine-Learning to Advance Basic Medical Research
Understanding how genes behave in different cells and tissue types throughout the body is breathtakingly complex, but it’s critically important to advancing our basic understanding of human biology. While basic-science researchers have become adept at identifying gene activity among large clusters of cells, pinpointing activity within individual cells has always been elusive. That is, until Moana tackled the challenge.
Moana is a machine-learning program that can infer the gene behavior of a lone cell by churning through massive databases of gene-related data extracted from larger cell clusters. The system (named after the favorite movie character of the daughter of one of the program’s developers) recently emerged from NYU Langone’s Institute for Computational Medicine.
The institute’s goal is to focus the power of advanced computing on basic medical research that’s becoming ever more dependent on enormous sets of gene and other data. “There are amazing discoveries to be made in new large-scale data sets,” says Itai Yanai, PhD, the institute’s inaugural director and professor of biochemistry and molecular pharmacology. “It’s become central to medicine, even if you can’t see it walking through the hospital corridors.”
Whereas researchers were previously limited to investigating the impact of a single gene at a time on disease or development, Dr. Yanai’s work has allowed sifting through whole strings of genes to pick out the most important patterns. “Not only can we look at all the genes,” he says, “but we can track them to an individual cell to see which particular genes are the most important.”
Machine learning is increasingly important to advancing the field, insists Dr. Yanai. “It’s becoming one of the keys to doing some of the most complex types of analysis on a ton of data,” he says.
But he’s also quick to point out that the biggest promise of artificial intelligence, or AI , in basic research is still floating out there in the future. “What we’d like to do is integrate AI into the discovery process, but we don’t know how to do that yet,” he says. “Can it figure out something new that we didn’t even know to look for? We don’t know yet if it can actually make novel discoveries.”
For now, basic-research discoveries will have to be made the old-fashioned way, but programs like Moana can at least crunch through the mountains of data that those new discoveries will undoubtedly spawn.