Photo credit: www.sciencedaily.com
Advancing Genetic Research for All: University of Florida’s Innovative Approach
Researchers at the University of Florida are tackling a significant issue in medical genetic research: ensuring it effectively represents and serves individuals from diverse backgrounds.
Under the direction of Kiley Graim, Ph.D., an assistant professor in the Department of Computer & Information Science & Engineering, the team’s work seeks to improve health outcomes by addressing “ancestral bias” within genetic datasets. This bias often stems from studies predominantly utilizing data from a single ancestral group, which limits the effectiveness of precision medicine and leaves many segments of the global population at a disadvantage in terms of disease prevention and treatment.
To combat this challenge, the research team has developed PhyloFrame, a machine-learning tool that incorporates artificial intelligence to factor in ancestral diversity in genetic information. With the backing of the National Institutes of Health, the aim is to enhance disease prediction, diagnosis, and therapeutic options for all individuals, irrespective of their ancestral backgrounds. The methodology and findings related to PhyloFrame were detailed in a recent publication in Nature Communications.
Graim’s pursuit of addressing ancestral bias was catalyzed by a dialogue with a physician, who expressed concern over the limited applicability of research to his varied patient demographics. This encounter inspired Graim to explore how artificial intelligence could close the gaps present in genetic research.
“I thought to myself, ‘I can fix that problem,'” said Graim, drawing on her expertise in machine learning and precision medicine, compounded by her training in population genomics. “If our training data doesn’t correspond to real-world data, we possess methodologies using machine learning to mitigate that issue. While these methods aren’t flawless, they hold significant potential to rectify this challenge.”
Utilizing information from the population genomics database gnomAD, PhyloFrame merges extensive collections of healthy human genomes with more specialized datasets concerning diseases that are used to train precision medicine models. This enhances the models’ ability to accommodate a variety of genetic backgrounds. For instance, PhyloFrame can differentiate between breast cancer subtypes, thus recommending optimal treatments tailored to each patient’s unique genetic profile.
The analysis of such vast data is no trivial task. The research team employs the University of Florida’s HiPerGator, recognized as one of the most formidable supercomputers in the nation, to process genomic data from millions of individuals—handling approximately 3 billion base pairs of DNA for each person.
“I didn’t anticipate the results we achieved,” Graim remarked, acknowledging the vital role played by her doctoral student, Leslie Smith, in the research. “What started as a modest project aimed at demonstrating the impact of incorporating population genomic data transformed into securing funding for developing more intricate models and refining the definitions of populations.”
What distinguishes PhyloFrame is its capability to maintain accuracy in predictions across diverse genetic backgrounds by recognizing genetic variations associated with ancestry. This attribute is particularly important as existing models are frequently built on datasets that fail to represent the full spectrum of the global population. A significant portion of the data currently available originates from research institutions and patients who have confidence in the healthcare system, thus excluding populations from rural areas or those with mistrust towards medical institutions. This omission complicates the development of universally effective treatments.
Graim estimates that approximately 97% of sequenced genetic samples derive from individuals of European ancestry, largely due to disparities in national and state funding priorities, as well as socioeconomic factors. For example, access to insurance can influence treatment opportunities, subsequently affecting the likelihood of receiving genomic sequencing.
“Countries such as China and Japan have recently made strides to close this discrepancy, resulting in an uptick in available data from these regions, yet it remains significantly less than that from Europe,” she explained. “Moreover, economically disadvantaged populations are often wholly excluded from these studies.”
Thus, Graim emphasizes the critical importance of diversity in training datasets.
“We aim for our models to be effective for each patient, not just those involved in our studies,” she asserted. “Having a diverse training dataset ultimately enhances the model’s efficacy for all populations, including Europeans. Utilizing population genomics data assists in preventing models from overfitting, thereby improving outcomes for everyone.”
Graim envisions that tools like PhyloFrame will eventually be integrated into clinical practice, supplanting conventional models to tailor treatment strategies based on individual genetic information. Future efforts will focus on refining PhyloFrame and broadening its applications to encompass additional diseases.
“My aspiration is to advance precision medicine by employing this kind of machine learning approach, enabling individuals to receive early diagnoses and personalized treatment plans with minimal side effects,” she stated. “Our goal is to ensure that the right treatment reaches the right individual at the right moment.”
The project has received funding through the AI2 Datathon grant award from the UF College of Medicine Office of Research, which is dedicated to empowering researchers and clinicians to utilize AI in enhancing human health.
Source
www.sciencedaily.com