Similarity measures for clustering SNP and epidemiological data

Abstract

The issue of suitable similarity measures for a joint consideration of so called SNP data and epidemiological variables arises from the GENICA (Interdisciplinary Study Group on Gene Environment Interaction and Breast Cancer in Germany) casecontrol study of sporadic breast cancer. The GENICA study aims to investigate the influence and interaction of single nucleotide polymorphic (SNP) loci and exogenous risk factors. A single nucleotide polymorphism is a point mutation that is present in at least 1 % of a population. SNPs are the most common form of human genetic variations. In particular, we consider 43 SNP loci in genes involved in the metabolism of hormones, xenobiotics and drugs as well as in the repair of DNA. Assuming that these single nucleotide changes may lead, for instance, to altered enzymes or to a reduced or enhanced amount of the original enzymes – with each alteration alone having minor effects – the aim is to detect combinations of SNPs that under certain environmental conditions increase the risk of sporadic breast cancer. The search for patterns in the present data set may be performed by a variety of clustering and classification approaches. I consider here the problem of suitable 2 measures of proximity of two variables or subjects as an indispensable basis for a further cluster analysis. In the present data situation these measures have to be able to handle different numbers and meaning of categories of nominal scaled data as well as data of different scales. Generally, clustering approaches are a useful tool to detect structures and to generate hypothesis about potential relationships in complex data situations. Searching for patterns in the data there are two possible objectives: the identification of groups of similar objects or subjects or the identification of groups of similar variables within the whole or within subpopulations. The different objectives imply different requirements on the measures of similarity. Comparing the individual genetic profiles as well as comparing the genetic information across subpopulations I discuss possible choices of similarity measures suitable for genetic and epidemiological data, in particular, measures based on the χ2-statistic, Flexible Matching Coefficients and combinations of similarity measures.