New statistical method for genetic studies could cut computation time from years to hours

March 17, 2010

In the ongoing quest to identify the genetic factors involved in disease, scientists have increasingly turned to genome-wide association studies, or GWAS, which enable the scanning of up to a million genetic markers in thousands of individuals.

These studies generally compare the frequency of genetic variants between two groups -- those with a particular disease and healthy individuals. Differences in the frequency of a given variant suggest the variant may be involved in the disease.

Over the last few years, such studies have successfully implicated hundreds of genes in human disease, and the research has been used to identify risk and protective factors for asthma, cancer, diabetes, heart disease, mental illness and other conditions.

But genome-wide association studies aren't perfect. In fact, the genealogy of study participants can sometimes prove a stumbling block to accurate findings.

"Unfortunately, differences in frequencies can arise for reasons unrelated to the disease if the individuals collected have ancestry from different regions of the world," said Eleazar Eskin, associate professor of computer science at the UCLA Henry Samueli School of Engineering and Applied Science, who holds a joint appointment in the department of human genetics at the David Geffen School of Medicine at UCLA.

"This problem, called 'population structure,' has led to many apparent discoveries of genes involved in disease which later turned out to be artifacts," he said.

In a new study to be published in the April edition of the journal Nature Genetics (currently available online), Eskin and his research group unveil a new computational strategy for GWAS that corrects for population structure and is both faster and easier to use.

One of the basic assumptions in typical GWAS is that participating individuals are "unrelated," and investigators typically perform screening procedures to ensure that pairs of individuals are not close relatives. However, due to the complex history of the human population, none of the individual pairs are perfectly unrelated, and each individual pair is somewhat distantly related to various degrees. This is referred to as "pairwise relatedness."

"Such a variety in degrees of relatedness -- which we call 'sample structure' -- can be manifested into two different forms: population structure and hidden relatedness. While typical statistical methods for GWAS handle only either of the two forms, our method can handle both aspects of sample structure simultaneously in a computationally efficient manner," said Hyun Min Kang, an assistant research professor in biostatistics at the University of Michigan and an author of the study.

"Moreover, if the samples come from a very homogeneous population, it is possible that some of the subjects are, in fact, distantly related," said Chiara Sabatti, professor of human genetics and statistics at UCLA and a corresponding author of the study. "In the analysis of GWAS, it is necessary to correct for such sample structure, which can lead to spurious association signals. The methods presented in our paper allow researchers to do this in a manner that is both fast and effective."

Eskin's team worked with a data set of 5,000 people from Finland who were born in the same year, tracked over an extensive amount of time, and had a large amount of population relatedness.

The 5,000 people produced a data set of 300,000 variants. From these 300,000 points of variation, the group examined pairwise relatedness between individuals, which means they compared the number of mutations each shared. From the mutations, Eskin's group could estimate how related individuals were to each other.

"It was very interesting to see how much these pairwise relations explained of the trait," Eskin said. "So what we did in this paper is we proposed a statistical method that also allowed us to correct for a wide range of sample structure by explicitly accounting for pairwise relatedness between individuals using high-density markers in modeling the distribution of observable traits."

This variance component in the new strategy, called EMMAX (Efficient Mixed Model Association Expedited), would capture the complex mixture of both population structure and hidden relatedness, direct byproducts of genealogy, and correct for these relationships when performing genetic mapping.

"Capitalizing on the characteristics of complex traits in humans, we made a few simplifying assumptions that allowed us to dramatically increase the speed of computations, making our approach readily applicable to genome-wide association studies with tens of thousands of samples," Eskin said.

"Our variance component model is actually a widely known classical model for genetic mapping," Kang said. "However it was too computationally costly to be applied to the current scale of GWAS involving thousands of individuals with hundreds of thousands of genetic variants because even the fastest method -- which we previously developed -- took years of computational time to analyze the data once. We further expedited the method by capitalizing the characteristics of most human association studies, reducing the computational time from years to hours."

According to Eskin, their method will also have a large impact on admixed populations, which are basically samples of individuals who have ancestry from multiple regions around the world. Studies on Los Angeles, for example, would benefit from this method greatly, as people in the city are very ethnically diverse and it's difficult to obtain very accurate estimates of people's ancestry.
-end-
The study was supported in part by the National Toxicology Program/National Institute of Environmental Health Sciences.

The UCLA Henry Samueli School of Engineering and Applied Science, established in 1945, offers 28 academic and professional degree programs, including an interdepartmental graduate degree program in biomedical engineering. Ranked among the top 10 engineering schools at public universities nationwide, the school is home to eight multimillion-dollar interdisciplinary research centers in wireless sensor systems, nanotechnology, nanomanufacturing and nanoelectronics, all funded by federal and private agencies.

For more news, visit the UCLA Newsroom and follow us on Twitter.

University of California - Los Angeles

Related Genetic Variants Articles from Brightsurf:

Researchers identify genetic variants linked to toxic side effects from bevacizumab
In the largest study of its kind, researchers have found two common genetic variants that can be used to predict whether or not cancer patients might suffer severe adverse side-effects, such as high blood pressure, from the drug bevacizumab.

Genetic risk of developing obesity is driven by variants that affect the brain
Some people are at higher risk of developing obesity because they possess genetic variants that affect how the brain processes sensory information and regulates feeding and behavior.

Genetic background influences disease risk from single-gene variants
Life can change dramatically when someone learns they are genetically predisposed to a disease.

Researchers identify novel genetic variants linked to type-2 diabetes
After examining the genes of more than 200,000 people all over the world who have type-2 diabetes, researchers from the Perelman School of Medicine at the University of Pennsylvania and the Veterans Health Administration's Corporal Michael J.

FSU researchers help discover new genetic variants that cause heart disease in infants
Florida State University researchers working in an international collaboration have identified new genetic variants that cause heart disease in infants, and their research has led to novel insights into the role of a protein that affects how the heart pumps blood.

Twenty four genetic variants linked to heightened womb cancer risk
Twenty four common variations in genes coding for cell growth and death, the processing of oestrogen, and gene control factors may be linked to a heightened risk of developing womb (endometrial) cancer, indicates the most comprehensive review of the published evidence so far in the Journal of Medical Genetics.

Genetic variants reduce risk of Alzheimer's disease
A DNA study of over 10,000 people by UCL scientists has identified a class of gene variants that appear to protect against Alzheimer's disease.

Rare genetic variants predispose to sudden cardiac death
By identifying rare DNA variants that substantially increase risk of sudden cardiac death, researchers have laid the foundation for efforts to identify individuals who could benefit from prevention strategies prior to experiencing symptoms.

Genetic variants for autism linked to higher rates of self-harm and childhood maltreatment
People with a higher genetic likelihood of autism are more likely to report higher childhood maltreatment, self-harm and suicidal thoughts according to a new study by researchers at the University of Cambridge.

Genetic variants with possible positive implications for lifestyle
A German and British research team lead by the Technical University of Munich (TUM) has examined the interplay between genetics, cardiovascular disease and educational attainment in a major population study.

Read More: Genetic Variants News and Genetic Variants Current Events
Brightsurf.com is a participant in the Amazon Services LLC Associates Program, an affiliate advertising program designed to provide a means for sites to earn advertising fees by advertising and linking to Amazon.com.