AI Tool Tackles Ancestral Bias in Genetic Research for Improved Precision Medicine

University of Florida develops PhyloFrame, a machine-learning tool designed to address ancestral bias in genetic data and improve precision medicine.

• New AI tool PhyloFrame addresses underrepresentation in genetic data.
• Tool significantly improves precision medicine predictions for diverse populations.
• Research highlights the necessity of diverse genomic data to enhance healthcare outcomes.

Researchers at the University of Florida have developed a new artificial intelligence (AI) tool, PhyloFrame, designed to tackle ancestral bias in genetic datasets. This development promises to significantly enhance precision medicine, ensuring that medical research and treatments better serve people from diverse ancestral backgrounds.

Addressing ancestral disparities

Ancestral bias in genomic data arises primarily because the majority of medical genetic research has traditionally focused on individuals of European descent. According to Dr Kiley Graim, an assistant professor in the Department of Computer & Information Science & Engineering and lead researcher of the study, around 97% of sequenced genomic samples originate from populations with European ancestry. This bias can severely limit the applicability and accuracy of precision medicine for underrepresented groups.

PhyloFrame employs advanced machine-learning techniques to integrate data from large-scale genomics databases, such as gnomAD, combining comprehensive databases of healthy human genomes with specific disease-related datasets. The approach significantly improves the accuracy of medical predictions across diverse ancestral groups, particularly benefiting populations traditionally underrepresented in genetic research.

“If our training data doesn’t match our real-world data, we have ways to deal with that using machine learning,” said Graim. “They’re not perfect, but they can do a lot to address the issue.”

Enhancing predictive accuracy

PhyloFrame was developed using computational resources provided by HiPerGator, one of the United States’ most powerful supercomputers. This enabled the team to process and analyse genomic information from millions of individuals, each comprising roughly 3 billion base pairs of DNA. Early results have demonstrated that PhyloFrame significantly enhances the accuracy of predictions for diseases, including identifying subtypes of breast cancer and suggesting appropriate treatments regardless of patient ancestry.

“I didn’t think it would work as well as it did,” Graim noted. “What started as a small project using a simple model to demonstrate the impact of incorporating population genomics data has evolved into securing funds to develop more sophisticated models.”

Graim emphasised the broader benefits of inclusive genomic datasets, noting that even populations well-represented in current datasets, such as Europeans, also see improved model performance when diverse data is used.

“Having diverse training data makes models better for Europeans, too,” she said. “It helps prevent models from overfitting, meaning they’ll work better for everyone.”

The team now aims to refine the capabilities of PhyloFrame further, extending its application to a wider range of diseases. Ultimately, Graim envisions that tools like PhyloFrame will be commonplace in clinical settings, providing tailored treatment plans based on individual genetic profiles.

“My dream is to help advance precision medicine through this kind of machine learning method, so people can get diagnosed early and are treated with what works specifically for them and with the fewest side effects,” she concluded. “Getting the right treatment to the right person at the right time is what we’re striving for.”

Graim’s research received funding from the UF College of Medicine Office of Research’s AI2 Datathon grant award.