Understanding Variations in DNA
By Shavantay Minnis
Raquel Assis, Ph.D., examines what happens to genes once they mutate. Do they evolve into traits like having red hair and freckles, or do they duplicate and delete to cause severe diseases, like cancer?
Using artificial intelligence (AI), Assis’ new approach trains machines to predict the outcomes of an altered gene. Her work creates a path for evolutionary biologists and other geneticists to develop better disease treatment approaches.
“Structural variations, or large-scale differences in our DNA, are key to both evolutionary adaptation and diseases like cancer or lung disease,” Assis said. “Our new machine learning methods will enable us to answer targeted questions about how structural variations drive evolutionary innovation.”
Assis, a faculty fellow in the Institute for Human Health and Disease Intervention, and associate professor in the College of Engineering and Computer Science, recently earned a $1.8 million grant from the National Institutes of Health (NIH), and a $596,571 grant from the National Science Foundation (NSF) for this and other projects. She will develop machine learning techniques that use genomic data to predict the short term and long-term evolutionary outcomes of DNA differences in both humans and animals.
Think of those who are lactose intolerant, she said, it began as a mutation in the genes, where a majority of the population had problems digesting milk. But once it evolved, individuals were able to continue drinking milk without any real harm. Yes, it’s still a mutation, just not one that creates major problems like diseases, she said.
“Defining a gene’s function is hard to quantify, but with our machine learning algorithms we can gain insight into how gene functions evolve after duplications, inversions, and other large-scale mutations,” she said. “These methods will also allow evolutionary biologists to answer an array of questions related to evolution across different species.”
When Assis began her research, she created computerized software programs to help researchers get a closer look at the connection between genes and traits. Her program could classify the functional outcomes of duplicate genes and other structural variations from patterns hidden in the genes.
She spent most of her efforts looking at the functional outcomes of large-scale variations or mutations like duplications, which create copies of existing genes. Such mutations can often lead to either an adaptation, such as a stronger immune response to a new pathogen, or a disease like cancer and diabetes.
“Because duplication is the most common type of structural variation observed in nature, much of my earlier work focused on developing approaches for learning about the evolution of duplicate genes,” Assis said. “Now I plan to design similar techniques for studying important classes of structural variations.”
When Assis entered college as a first-generation student at the University of Florida, she started her studies in biology, where she learned about the evolution of genomes, which is the complete blueprint of all the individual genes in the human body. Her interest in the genome and other neuroscience courses led her to complete a bachelor’s degree in psychology and zoology in 2006, she said.
Her time as an undergraduate also led her to questions like “why is there so much variation within genes” and “what kind of software programs could study it,” she said. Realizing that these programs were not created yet, during her doctorate program at the University of Michigan in 2007, Assis began studying bioinformatics, a field that combines computer science and biology to study genomes. There she created her first computer software program called Bridges, which allowed her to identify and search for similarities in the DNA of one human to the next.
Once she earned her doctoral degree in 2011, Assis continued her study of genomes as a NIH postdoctoral fellow at the University of California, Berkeley. During her time there she designed the first method for predicting the evolutionary outcomes of duplicate genes.
When she completed her NIH postdoctoral fellowship in 2013, the following year she started a research and teaching position as an assistant professor in the department of biology at Pennsylvania State University.
In 2019, she joined FAU and immediately continued her work in bioinformatics creating other methods to study structural variations. Now with a five-year NIH Maximizing Investigators’ Research Award and three-year NSF grant, Assis plans to further her work using machine learning to study the evolution of genes and determine other questions like how natural selection shapes the evolutionary trajectories of structural variations,
she said.
“What I enjoy most about my research on structural variations is that it has the potential to create a bigger picture for scientists to view human diseases,” Assis said. “And much like our genes, this picture continues to evolve.”