In our research, we have three groups of genotypes: 1) Resistant, 2) Susceptible and 3) Unknown (maybe resistant, maybe susceptible). We performed NBS (Nucleotide Binding Site) profiling in all genotypes and the results are ready as 0/1 scores. Thus, we want to find a way to make a correlation between the presence/absence of bands and the resistance/susceptibility of genotypes to guess the resistance status in unknown genotypes. I would be grateful if you could kindly share your experiences in this context.
If you want to have a measure of the "strength" of the relationships between those categorical phenotypes and genotypes, you may use the chi square formula in contingency tables. Dividing by n and taking the square root, you can get the equivalent to the correlation coefficient for a 2 x 2 contingency table (in your case, it is 3 x 2). You can find the definition of such a correlation coefficient in Analysisi of Human Genetic Linkage(Ott, 1999, page 282).
I understand that you want to predict the resistance status in unknown genotypes. There are nice and powerful machine learning techniques for that. You can start with the k-nearest neighbors method, or explore the random forest technique.
Best wishes,
Humberto
Dear Ali
In this case, you have categories, So you can use Pearson's chi-squared test (χ2), which can be used for unpaired data from large samples, and there are many softwares can help you to do the test, and I prefer SPSS.
Good Luck
Dear Ali,
I suppose your question is actually two questions: 1 - which NBS are specific to Resistant/susceptible genotypes and 2 - is actually the question you have sent.
Besides three more things to be considered, to my point of view: do the biological forms you test are genetically homogeneous or not; how did you characterize the phenotype (just resistant / susceptible or there are several traits for each); and are the NBS are reproducible?
I used to do similar things with DNA polymorphism in association to resistance. I worked with highly inbred lines and their hybrids. Any DNA band I obtained on electrophoresis had to have frequency of at least 95%.
The first you may do is to identify really specific NBS associated just with one of the genotypes you've been testing and extract them from further analysis as genotypes- specific. For the rest of the 1/0 profiles of NBS you perform cluster analysis by e.g. Neighbour_joining method based on matrix of genetic distance/genetic similarity calculation as 2N/(N1+N2), where N - is number of similar NBS between two genotypes, N1 - total number of bands of genotype 1 and N2 - of genotype 2.
Another way is to use Pearson/Spearman correlation to generate the matrix and use it to construct the clusters.
You can use any statistical software to generate the clusters based of genetic distance/similarity and evaluate weather your genotypes are grouping in clusters according to the expected genotypes. As you possibly understand it is important to exclude genotype-specific NBS as they will possibly place these genotypes in distant clusters.
Then you need to analyze the tree in order to identify the level of similarity/distance/correlation which gives you best way to obtain the clusters you expect according to your phenotypes.
Once you will repeat/generate more experiments you will accumulate data in order to obtain statistical significance.
Good luck
Dear Dr. Levitchi,
Thank you for your helpful discussion.
I used biological replications and the NBS bands are reproducible.
For phenotypic observations, I have only resistant/susceptible traits.
Do you recommend the Mantel test?
- Badges
- Science topic
- Similar topics
- Agricultural Science
- Agronomy