Professional Documents
Culture Documents
The Meaning of Interaction
The Meaning of Interaction
The Meaning of Interaction
net/publication/49676026
CITATIONS READS
98 700
3 authors, including:
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Robert C Elston on 29 May 2014.
I II
Additive Statistical interaction measured on an additive scale, In some contexts, like evaluating the efficacy of two
interaction i.e. the combined effect (risk) of the factors is higher or medicines, the combined effect is equal to the sum of
less than the addition of the individual main effects. the effects of the medicines given separately.
Epistatic One gene (or any genetic factor) masks or suppresses Like ‘epistasis’, broadly implies any type of statistical/
interaction the effect/action of other(s). physical interactions between genetic factors.
Intragenic Interaction between different SNPs within a gene. Interaction between alleles at the same locus
interaction (dominance).
Multiplicative Statistical interaction assessed on a multiplicative scale. The product of main effects.
interaction
Physiological Cheverud and Routman called interaction for A biological process/mechanism similar to physical
epistasis/interaction unweighted means ‘physiological epistasis’. interaction.
Quantitative The magnitude of the effects of one factor varies Synonymous with ‘statistical interaction.’
interaction across the levels of another, but not the direction
(removable/non-crossover interaction), contrasted
with ‘qualitative interaction’.
Synergistic/ The combined effect of the factors is greater/less than The combined effect of the factors is greater/less than
antagonistic the sum of the individual effects. any of the individual effects.
interaction
Statistical Contrasted with ‘physical interaction’, Contrasted with ‘compositional epistasis’ [27] with a
epistasis/interaction merely emphasizes its mathematical nature. special emphasis on its population-average property.
ambiguity, it is generally preferable to couple the term the actual gene. In genome-wide association studies
‘interaction’ with other descriptive words or phrases, al- (GWAS), routine analyses are mostly based on SNPs
though many of these still have no clear-cut and gener- where a gene locus may involve hundreds or thousands of
ally accepted definitions (see table 1 for examples). The SNPs. The number of possible SNP pairs (let alone SNP
increased complexity of studies in human genetics makes trios, quartets, etc.) grows very rapidly with the number
them of necessity collaborative in nature, and collaborat- of genes – or SNPs in each gene – selected, posing a com-
ing researchers should always be crystal clear about the putational challenge and even greater difficulty in inter-
exact meaning of the words they use, because the same pretation. Novel methods for tests at the gene or SNP set
term (such as ‘gene-gene interaction’ and the terms listed level are becoming available but are only beginning to be
in table 1) can convey quite different meanings to people used in GWAS. To answer the question from the physical
with different scientific backgrounds. or molecular standpoint, it is essential to ask a more basic
question that we often neglect: what is a gene? The central
dogma states that genetic information flows from DNA to
What Is Gene-Gene Interaction? RNA to protein. But whereas 95% of all DNA is tran-
scribed to RNA, very little of this is translated to protein.
Is gene-gene interaction the interaction between genes? Physical interaction can be triggered at any stage. As con-
From the standpoint of population genetics or genetic ep- ceptually illustrated in figure 1, all the entities coexist and
idemiology, a frequently used definition of gene-gene in- coordinate to form a dynamic cellular system. Two DNA
teraction is the interaction between alleles at different loci. variations may interact with each other directly, though,
The term ‘locus’ is now being used not only for the loca- as far as is known, not very commonly. If we define a gene
tion of a gene, but also (unfortunately, perhaps) for the as any stretch of DNA with function [4], it is mostly gene
location of any genetic variant or marker that is nearby or products, not the genes themselves, that interact physi-
within a gene, which can serve as a surrogate for studying cally. DNA is packaged into chromatin fibers, which oc-
cupy distinct areas in the nucleus called ‘chromosome ter- There is still much confusion between ‘statistical gene-
ritories’ (CTs). Some regions of high gene density may gene interaction’ and ‘biological gene-gene interaction’,
loop out of CTs and move to contact other sites far from partly because of the use of the word ‘epistasis’ (epi =
the region, causing an interesting event termed ‘gene kiss- upon, stasis = stand). This word was coined over a cen-
ing’ [5]. DNA can be specifically recognized by proteins tury ago by Bateson [6], the same person who suggested
with special structures (DNA binding motifs) to form the word ‘genetics’. He used the word to describe only the
protein-DNA complexes, which regulate both DNA repli- masking action whereby an allele at a locus suppresses the
cation and gene expression. Current knowledge about effect of an allele at another locus, which results in a de-
protein-RNA interactions, which are interactions be- parture from the expected dihybrid ratio 9: 3:3: 1. The
tween gene products, remains limited, but they are gener- term later acquired a much broader meaning that is al-
ally believed to be involved in RNA metabolism and most synonymous with gene-gene interaction as we have
translation. Considerable, perhaps disproportionate, at- defined it. Fisher [7], in his seminal paper of 1918, used
tention has been given to the study of the interactions be- another noun form (‘epistacy’) to describe the deviations
tween proteins, including enzymes and other functional from the additive effects of alleles at different loci. This
molecules that control various metabolic and regulatory term was soon replaced by ‘epistasis’ in the quantitative
pathways. This is largely driven by the availability of high- genetics literature [8]. The meaning of interaction in ge-
throughput proteomic techniques. Physical gene-gene in- netics has now evolved into two rather divergent direc-
teraction occurs relatively rarely as a fraction of all the tions. ‘Interaction’ (or ‘statistical interaction’) is used by
physical events that jointly involve genes and their prod- statisticians to describe the non-additivity in generalized
ucts, and these events occur irrespective of whether they linear models. This definition is thus close to Fisher’s
have any effect on the final phenotypes. In the light of the term ‘epistacy’. Biologists use the term ‘biological interac-
modern definition of a gene [4] where final function is tion’, or simply ‘interaction’ to mean the joint action of
emphasized, it is irrational to give priority to protein-pro- two or more factors, whether or not an additive statistical
tein interactions rather than RNA-protein interactions. model is sufficient, thinking of the physical interaction
For what follows we explicitly (rather than implicitly, as is between molecules. We suggest that: (1) unless the mech-
so often the case) define the term ‘gene-gene interaction’ anism is known, the term ‘biological interaction’ is better
to include gene-gene product and gene product-gene replaced by the less specific term ‘joint action’; (2) ‘inter-
product interactions. action’, if quantified, should be based on statistical con-
3.1 E2
E2
E1 1.76 E2
1 E1 0.95
0.9 E1
b G1 G2 c G1 G2
b G1 G2
G1
E2
G2
E1
Fig. 3. Removable interaction. Non-parallel lines in a and b indi-
cate removable interactions. Because the slope of the upper line is
greater than the slope of the lower line, a is often referred to as
synergistic interaction. Interactions in a and b, where the lines d E1 E2 e G1 G2
have slopes of the same sign and do not cross throughout the rang-
es considered, can be removed by changing the response scale.
E2
E1
size of the study, because the data obtained are observa- weights over all cells being equal to unity). Mathemati-
tional rather than from a controlled experiment; but they cally, letting wij be the proportion of the population in cell
may also be caused by selection against certain genotypes. ij, we define the weighted grand and marginal means re-
Because of this, when analyzing data we need to concern spectively as
ourselves with another factor that has not been given
wijij , i. wijij / wij and . j wijij / wij ;
proper attention – the unequal numbers of observations i j j j i i
in cells, known generally in statistics as unbalanced data.
Unbalanced data, like empty cells, can occur either unweighted means correspond to all the wij being equal.
due to chance alone or due to unequal population propor- With these weighted means, the contrasts and tests for
tions, which are then true population parameters. When main and interaction effects can be set up as in equations
not due to sampling variation, such proportions corre- (1) and (2). When the main effects are defined using ei-
spond to allele/genotype frequencies (exposure frequen- ther uniform weights (e.g., wij = 0.25 for all four cells of a
cies) in the case of a gene-gene (gene-environment) study, 2 ! 2 table) or proportional (meaning that
and their proportions – model parameters – may vary wij wi.w. j , or wiiw jj / wijw ji 1 ,
across different populations. When testing for main ef-
fects, one has a choice between contrasting unweighted corresponding to weights that have an odds ratio (OR) of
means or weighted means. Unweighted (equally weight- 1 in a 2 ! 2 table), the results are the same. But when the
ed) means are computed as simple averages of the cell weights are unequal and disproportionate (fig. 5), an ap-
means, while weighted means, if we use a common type parent interaction may result, even if ij + jj = ij + ji.
of analysis, weights the cell means by the cell sizes or pro- Thus, reverting now to population parameters rather
portions. We now focus for the moment on the effect of than sample estimates, the very meaning of interaction
these two different ways of defining main effects on the model parameters depends on how the main effects are
estimates of interaction found in samples, using the sym- defined. In the analysis of variance, this is referred to as
bol wij for the weight given to cell ij (with the sum of the being due to a non-orthogonal design where the interac-
tive and dominance effects, based on unweighted aver- who has his head in the oven and feet in the freezer.
ages of genotypic values, which, unlike the traditional
‘main effects’ used by statisticians, are independent of ge-
notypic frequencies in the population. Gametic Phase Disequilibrium and Interaction
All analyses of statistical interactions are model de-
pendent. An apparent departure from additivity may ‘Linkage disequilibrium’ (LD) is another widely, but
merely be an artifact of the particular measurement scale, often misleadingly, used term in genetics. It is semanti-
may reflect non-proportional sub-population probabili- cally impossible for unlinked loci to be in LD, but un-