Scientists at Roche have improved upon a computational method they developed several years ago to identify disease-causing genes in the mouse genome.
In a paper published in the Oct. 22 issue of Science, the scientists describe the method, which relies on information in the company’s freely available mouse SNP database (http://mousesnp.roche.com/). At the same time, Roche has released a haplotype map for 16 Mus Musculus strains through the database.
Gary Peltz, head of genetics and genomics at Roche Palo Alto, told BioInform that the company’s new approach offers a thousand-fold improvement in precision over a similar method that it first described in Science in 2001 [BioInform 06-11-01].
As described in the 2001 paper, the Roche method — called Digital Disease — predicted large genomic regions in which a disease trait was likely to appear, but these regions could contain up to hundreds of genes. Peltz said the new method — dubbed HapMapper — is now precise enough to home in on a single gene, or even a subgenic region, implicated in a disease.
The key to the improvement, Peltz said, is the haplotype patterns in the mouse genome. “In our 2001 paper we artificially divided the genome into 60-megabase segments for computational analysis. Here, we’re taking advantage of block-like structures that naturally exist in the genome of the inbred strains. Identification of these structures helped quite a bit with the mapping,” he said. The haplotype blocks average only around 30 kilobases in size, Peltz said, which greatly improves the accuracy of the approach.
In addition, Peltz said, the new method relies on much more stringent statistical techniques. “In the initial paper, in 2001, we had a relative statistic — there was no ability to assess the predictions in an absolute sense. The correlations between phenotypic traits and genetic variation placed in rank order.” Now, the Roche team uses absolute statistical criteria that provide a p-value to assess the predictions.
“We measure differences in any selected parameter among the inbred strains,” he explained. “You order the selected strains, make the desired measurements, and then we use our method to answer the following question: ‘Given this distribution of genetic variation amongst the strains, where is [the] gene with the best correlation with this pattern of trait variation?’”
A process that would take years to carry out by conventional genetic analysis “can literally be done in an afternoon, once the phenotypic data is available.”
In last week’s Science paper, the Roche team described how they used the method to identify genetic loci that regulate differential response in a metabolic response pathway among the inbred strains. Peltz said that based on this work, Roche is “examining a wide range of drugs that are of relevance, of biomedical importance, to try and better understand their metabolism and their mechanism of action.”
Peltz cautioned that the haplotype-based computational approach works well for inbred mice, but wouldn’t be useful for humans because of the “confounding problems” associated with human haplotypes. “The mouse is much easier to analyze because [you are] examining an organism with only one allele for each gene. You’ve eliminated the effects of heterozygosity. In the mouse, you don’t have to infer anything; you know the haplotype. In humans, you have to infer the haplotype.”
“Our approach is to use the mouse, identify the key genes and pathways, and then go study them more specifically in the human, rather than try to sort it out in the human where it is much more complicated.”
So far, he said, the Roche team has analyzed the pattern of genetic variation in 200 to 300 genes in the mouse that are relevant for drug metabolism. “We think that this method can be used to study drugs that have an unknown mechanism of action, or a mechanism of toxicity. We can use this method to help figure out how these drugs act,” he said.
The Roche group is also using the computational approach in an NIH-funded project to study mouse pharmacogenetics. In this project, the team will measure pharmacokinetic parameters for at least four different therapeutic agents across 11 inbred mouse strains, and will then use HapMapper to identify the genetic loci responsible for the variation in pharmacokinetic responses.
“The goal is that with that understanding ... you can reduce the number of drug candidates that potentially fail in later stages of development,” Peltz said.
Peltz and his colleagues have further enhancements planned for HapMapper. While he did not provide much detail, he said the group may even be able to improve upon the haplotype structures that enabled the current version of the system. “We are looking at a new way of characterizing the pattern of genetic variation that will make computational mapping even more feasible,” he said. “We think that in the near future we can even get away from these rigid block-like structures and get into a different type of structure that we can use for genome analysis.”