Configural Frequency Analysis as a Method for Genetic Association Studies in Complex Diseases

Many complex diseases carry a genetic component, and in most cases that component does not reduce to a single faulty gene. What seems to matter instead is how several polymorphic loci act together. If that assumption holds, the right unit of study is not one gene at a time but the combination of alleles that patients tend to carry and that separates them from a comparable group of unaffected people.

This paper puts forward a non-parametric way of handling that multivariate problem. The tool is Configural Frequency Analysis (CFA), applied to genetic association data. Rather than testing individual markers, CFA treats each possible configuration of alleles as an object of interest and asks whether it turns up more or less often than chance would predict among affected individuals.

Two kinds of result come out of this. A configuration counted more often than expected (f greater than e) is called typical; one counted less often than expected (f less than e) is called anti-typical. In the context of candidate genes, these configurations are haplotype allele combinations, so the analysis flags which specific arrangements cluster in patients and which are relatively avoided.

The approach fits both of the usual study designs. In a case-control setting, the observed frequencies of haplotype configurations among cases are weighed against the expected frequencies. Those expected values come from the control sample, which is taken to reflect how the alleles are distributed across the population at large.

Family-based designs supply their control in a different way. Here the alleles that parents did not pass to the affected offspring stand in as the comparison set. The transmitted configurations are then judged against this non-transmitted background, which sidesteps some of the population-stratification worries that haunt case-control work.

To show the method at work, the paper draws on clinical data published by Hoehe and coworkers in 2000. Running CFA on that material illustrates what the procedure produces in practice, and it gives a concrete basis for weighing where the method helps and where it falls short.

The closing discussion takes up those strengths and weaknesses directly, so that readers can see the conditions under which a case-oriented, configuration-level analysis adds something that gene-by-gene testing misses.