GeneSIS: enhancing transferability of polygenic scores with variant-level gene-by-sex interaction effects
The GeneSIS framework enhances the transferability of polygenic scores across diverse genetic ancestry groups by integrating variant-level gene-by-sex interaction effects, resulting in significantly improved disease prediction accuracy for non-European populations.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer
For decades, the promise of precision medicine has been to tailor healthcare to the unique genetic makeup of each individual. The idea is that by reading a person's DNA, doctors could predict their risk for diseases like heart disease or diabetes long before symptoms appear. This prediction relies on a tool called a polygenic score, which acts like a genetic report card. It sums up the tiny effects of thousands of genetic variations scattered across a person's genome to estimate their likelihood of developing a specific trait. However, a significant problem has emerged: these genetic report cards work remarkably well for people of European ancestry but often fail for everyone else. The scores lose their accuracy when applied to people with African, South Asian, or other genetic backgrounds, creating a dangerous gap in healthcare equity. This failure happens partly because the genetic patterns in different populations are not identical, and partly because the standard models used to build these scores assume that genes act in a simple, straight-line fashion, ignoring the complex ways they might interact with a person's environment or biology.
Researchers have long suspected that the missing piece of the puzzle lies in these ignored interactions. Specifically, they wondered if the effect of a gene might change depending on whether a person is male or female, a phenomenon known as a gene-by-sex interaction. While scientists have known for a long time that men and women can have different biological responses to the same genetic variants, most predictive models have treated everyone the same, averaging out these differences. A new study by Yosuke Tanigawa and Manolis Kellis introduces a fresh approach called GeneSIS, which stands for GENE and Sex Interaction Score. Instead of forcing all genetic data into a single, rigid mold, this new framework allows the model to learn directly from the data how genes behave differently in men and women. By analyzing the genetic information of nearly 407,000 diverse individuals from the UK Biobank, the team built a system that can detect these subtle, context-dependent effects and use them to improve predictions.
The researchers applied this new method to 99 different traits, ranging from blood pressure and cholesterol levels to body measurements like hip circumference and body mass index. They compared their new GeneSIS models against the standard, linear-only models that have been used for years. The results showed that the new approach successfully identified a specific set of genetic variants where the effect size or direction changed between sexes. In the final models, about 8% of the selected genetic variables were these sex-specific interactions, a finding that was rigorously checked and confirmed by comparing the results against separate analyses of men and women. This validation step ensured that the model wasn't just finding random noise, but was actually capturing real biological differences in how genes function in males versus females.
The most striking outcome of the study was a dramatic improvement in prediction accuracy for people of African ancestry, a group that has historically been underserved by genetic research. For a trait like hip circumference, which is a key measure of obesity and metabolic health, the new model performed significantly better than any previous method. In the test group of African individuals, the new model's ability to predict hip circumference was nearly four times better than the standard linear model. This improvement was not a fluke; it held true across a wide range of traits and was statistically robust. The researchers found that while the standard models often struggled to transfer their predictions from European populations to African populations, the GeneSIS models, by accounting for these sex-specific nuances, were able to bridge that gap. The improvement was so substantial that for hip circumference in African individuals, the new model outperformed every other publicly available prediction tool currently in existence.
Beyond the numbers, the study offered a glimpse into the biological mechanisms driving these improvements. The researchers looked closely at the specific genes the model selected and found that they pointed to plausible biological stories. For instance, the model highlighted a variant in a gene called GCKR, which is known to regulate glucose metabolism. The study showed that this variant has a complex relationship with body fat, but its effect is different in men and women, and it is also linked to other factors like menopause timing and hormone levels. This suggests that the model is not just finding statistical patterns, but is uncovering real, biologically meaningful interactions. Furthermore, the study found that the genes involved in these sex-specific effects were enriched for pathways related to inflammation and lipid metabolism, processes that are central to obesity and heart disease.
The implications of this work extend beyond just better prediction scores. By demonstrating that context-dependent effects, such as the interaction between genes and sex, are crucial for accurate prediction, the study challenges the long-held assumption that a single, universal genetic model can work for everyone. The researchers showed that by embracing the complexity of how genes interact with the body's context, they could create models that are more inclusive and more accurate. This is particularly vital for populations that have been left behind by previous methods. While the study focused on sex as the primary context, the framework is flexible enough to potentially incorporate other environmental factors in the future. The success of GeneSIS suggests that the path forward for precision medicine lies not in simplifying our understanding of genetics, but in building models sophisticated enough to capture the full, intricate reality of how our DNA works in the diverse tapestry of human life.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.