Introductory statistical genetics for plant breeding 2

Published on April 27, 2016   62 min

Other Talks in the Series: Statistical Genetics

0:04
An overview of topics that we will be dealing with, we continue in deeper view on what genotype to phenotype models are, and then we get to examples of these genotype to phenotype models. We will talk about QTL mapping, QTL mapping strategies as an example of how to use genotype to phenotype models. We will look at an analysis of genotype-environment interaction via a model that is called Finlay Wilkinson model, Finlay Wilkinson regression. And then we will finish with an example of what we call multi environment trial QTL mapping in which we use rather a advanced statistical model to find out which g10 QTLs are driving our phenotypic variation.
0:45
Okay, let's start with our next section, in a sense this is the most important section, I think, of the talk. I'm going to talk in more detail about a number of classes of genotype to phenotype models.
0:58
So on this slide, at the bottom, we see a model formula in which the genotypic main effect is covered in red and the other terms refer, let's say, to the response that we can observe for a particular plot in an environment j or trial j or a genotype i. And then at each of these trials, we also have a randomized complete block design, so we have an extra term called bk(j) in which, let's say, there is a block that is nested within a trial. So if we take the whole of this formula, we have an interceptor Mu, we have the environmental main effect, let's say, a trial intercept, we have an effect for the block inside the trial, we have genetic effect, and we have a final error term. And for breeding purposes, it's specially this genetic main effect which is important. Now in this case, on purpose, I did not include a specific term for the genotype-environment interaction, so idea is here that we have a number of trials that will allow us to make selection out of a number of genotypes. And we focus, in this case, on the average effect across those different trials. The formula is also stated in words, you can read at yourself. Phenotype = Environment + Block within Environment + Genotype + Error. And so typically, if breeders perform a set of trials, there after selecting the genotype with the best underlying genetic effect, this is also called the breeding value in many contexts.
2:30
Now, we see two, what we call recent modifications to this original scheme of phenotypic selection and let's say, since 25 years, we have access to molecular markers. And so on the top part of the slide, you see that the genotypic main effect of the former slide is replaced by regression on a number of markers, and in this sense, we can think of a QTL model, we will see more about these kinds of model in a moment. But we can think of a QTL model that approximates the original genotypic main effect. In this formula that we see at under markers assisted selection, we see that we have a number of Q, QTLs with QTL substitution, allele substation effects aq, and the sum of all QTL effects gives us an approximation to the original genotypic value of the former slide. There is also a residual and this residual is indicated by gi*, and of course, if the QTL mapping is successful, then the majority of the genetic differences can be explained by the QTLs and selection on QTL effects will be more or less equivalent to selection on the genetic values. Of course, the idea is that these markers are available at an early stage. So if we know the marker effects from any kind of training sets, let's say, we have done into past QTL analysis that allow us to estimate the QTL allele substitution effects if we have those QTL effects, if we have estimates for them, then we can use them to predict what a new genotype will do in certain conditions. We don't have to perform the field trials. We can also predict the value of the genotype by just looking at the markers. At the bottom part of the slides, we go slightly further. Since five years, a number of plants species can be sequenced. This has led to a full genome representation of the DNA variation and we have that in principle, we should have access to also the causal variants that underlie phenotypic variation. So at the bottom, what we tried to do is we replaced the original genetic differences by regression or old markers that we have or the full sequence information and this should provide us a rather close approximation to the original genetic differences. And so in this case, having access to the markers scores or the marker alleles, the marker genotype, will allow us to give very accurate prediction or the genetic differences which is a big help in selecting between promising genotypes.
Hide

Introductory statistical genetics for plant breeding 2

Embed in course/own notes