November 18, 2010

Joint effects explain some hidden variance (Culverhouse et al. 2010)

It seems that my lego-block paradigm has found support after all. The authors found genetic effects for their trait of interest (nicotine dependence) for pairs of SNPs, where the SNPs themselves showed no individual effect. Thus: unremarkable "commodity" building blocks combined to produce a particular effect.

Note that this was done on only pairs of SNPs. But, there is no reason to think that there won't be triads, or tetrads, or n-nads of SNPs having such effect, and the important thing is: if n-1 SNPs have no effect, n might. To give an everyday analogy: press Ctrl: nothing happens; press Alt: nothing happens, press Del: nothing happens, or perhaps a character is deleted, but press them all together, and all of the sudden something big happens.

There is a catch, however: for independent SNPs, the number of individuals that possess a particular n-long combination decreases exponentially with n. In short, you'd need to sample the whole population of the Earth, and you'd still not be able to find any individuals having some particular effective n-long combination, let alone a large enough sample to establish a statistical dependence with the trait of interest.

To reiterate: genome-wide association studies treat humans like black boxes: flip a SNP and see if the person is nicotine dependent or not. Or, as in this study, flip two SNPs that looked like "dead switches" when you tried to flip them individually. That approach is a dead end for most complex traits of interest, and the way forward is to get into the box, and see what genes actually do.

HUMAN GENETICS DOI: 10.1007/s00439-010-0911-7

Uncovering hidden variance: pair-wise SNP analysis accounts for additional variance in nicotine dependence

Robert C. Culverhouse et al.


Results from genome-wide association studies of complex traits account for only a modest proportion of the trait variance predicted to be due to genetics. We hypothesize that joint analysis of polymorphisms may account for more variance. We evaluated this hypothesis on a case–control smoking phenotype by examining pairs of nicotinic receptor single-nucleotide polymorphisms (SNPs) using the Restricted Partition Method (RPM) on data from the Collaborative Genetic Study of Nicotine Dependence (COGEND). We found evidence of joint effects that increase explained variance. Four signals identified in COGEND were testable in independent American Cancer Society (ACS) data, and three of the four signals replicated. Our results highlight two important lessons: joint effects that increase the explained variance are not limited to loci displaying substantial main effects, and joint effects need not display a significant interaction term in a logistic regression model. These results suggest that the joint analyses of variants may indeed account for part of the genetic variance left unexplained by single SNP analyses. Methodologies that limit analyses of joint effects to variants that demonstrate association in single SNP analyses, or require a significant interaction term, will likely miss important joint effects.



Anonymous said...

I guess the short answer is that everyone is looking for the easy way out using SNPs instead of studying actual genes and finding out their effects, and what modifies their action. Association studies of SNPs with disease outcomes is not enough. I find it very hard to believe that a SNP or even a group of SNPs whether within a gene or outside a gene or on the opposite dna strand from the gene has much bearing on disease outcomes. At its worse, the SNP may cause transcription errors so that a different amino acid is produced which may alter the effectiveness of the protein the gene produces but their must be more in the etiology of diseases than having on codon misplaced in an otherwise huge protein produced by a gene.

I am very skeptical on this matter and consider what 23andMe and the others do, voodoo medicine.

Andrew Oh-Willeke said...

Single SNP's that are part of a joint effect should show a statistically significant relationship to the trait in question, just muted from the entire effect because the SNP will produce an effect when its partners are present. So, a lot of the time, one can look for joint effect simply by setting the statistical significance threshold for SNPs fairly low, to reduce the number of potentially relevant points, and then analysing the dramatically reduced data set for joint effects, starting with the SNPs that have the most significant relationships.

This should find most joint effects except for those that are positive in connection with one SNP, but negative in connection with another SNP (think of it as a power booster SNP), producing no overall effect from that SNP when averaged out.