An interview with Erling Mellerup: The importance of combinatorial analysis

June 06, 2013 − by Suzanne Elvidge − in GenoKey people, GenoKey's big data analyses − No Comments

Many diseases, such as cancers, are caused by complex interactions of genetic changes and biological and lifestyle factors. However, a lot of research projects, despite analysis of thousands of genetic changes, only focus on single locus  strategies in the statistical analysis of the results, which may tell just part of the story. In an interview, Erling Mellerup, Associate Professor, Department of Neuroscience and Pharmacology at the University of Copenhagen, explains the importance of combinatorial analysis of the genetic changes using his work in bipolar disorder as an example.

Q. How did your collaboration with GenoKey begin?

A. We were looking at genetic changes in a study of bipolar disorder, and want to find out the influence of combinations of genetic changes on the disease. We analyzed 803 SNPs [single nucleotide polymorphisms], and then combined the SNP genotypes into groups of three. This, theoretically, will give 2,321,319,627 combinations.

Initially we used a data mining program called Clementine, but this wasn’t suitable for the scale we were working at – we simply had  simply too much data. I was almost ready to give up, and then I heard about the work Gert Møller was doing at Array Technology. This included software that solves complex configuration problems quickly, but using only small amounts of computer memory.

Q. How did combinatorial analysis help your research?

A. Using GenoKey’s technology, we were able to count the number of  combinations in the 1355 healthy people and 607 patients with bipolar disorder. This took only a few minutes, and we found a total of 1,985,613,.130 SNP genotype combinations, relatively close to the theoretical maximum.

Our next step was to work out which combinations are important. Of the 1.985 billion combinations seen over all, we found that about 1.7 billion combinations were seen in both controls and patients, with 200 million combinations in the control group only and 58 million in the patient group only. Breaking this down further, we found that just 1180 combinations were seen in groups of nine or more patients. Therefore, using GenoKey’s process reduced the number of combinations from nearly two billion to just over a thousand, a number that was much easier to analyse.

Q. What did you do next – 1180 is still a lot of potentially important combinations to analyse

A. We looked closely at this smaller group of combinations, and found four clusters of around 45 combinations of three SNP genotypes, each containing around 40 patients. There was almost no overlap between the patients in the four clusters. While the single SNPs and the single combinations of three SNP genotypes had no statistically significant link with bipolar disorder, the clusters were highly significantly linked with the disease.

Each patient has his or her own personal pattern of SNP genotypes, unique to each patient but broadly similar within each cluster, and all linked with signal transduction pathways.

Q. That’s an interesting result, but how is it relevant to doctors and patients?

A. Each  person has a personal pattern of genotypes and of clinical data, and because of this, we were able to link the genetic data in the clusters with clinical symptoms. We found statistically significant links between the clusters of SNPs and the symptoms, such as manic and depressive episodes. This could allow doctors to subdivide patients into subtypes of bipolar disorders, and tailor treatment to both their symptoms and their genetics.

Q. Could combinatorial analysis work in other disorders and with other kinds of data?

A. We believe so. We are currently looking at SNPs and clinical data from patients with neuroblastoma, and we are seeing significant clusters of SNP combinations. We are eager to try the technology in more diseases. GenoKey’s technology isn’t just limited to combinations of three SNPs – it can handle larger groups of SNPs creating many more combinations. The technique also has potential with any type of data, as long as it can be turned into figures.

Post a Comment

Your email address will not be published. Required fields are marked *