Hallo,
I compare SNPs to the clinical effictivness of a drug. Actually i know that my samplesize is not so huge.
I performed fisher´s exact test to find out if there is a correlation between beeing nonresponder or responder to the drug. At the second step i wanted to proof this by using a binary regression. The problem is i get different results. At one point i have no signicance in fisher but in logistic regression. How would you interprate this?
Greetings and thanks for the help
Tristan
Dear Tristan Klintz ,
Thanks for the question.
To begin with, these are two different tests, with very different assumptions and computation of p-values. Hence, they should yield different results.
However, what I want to point out is your decision on the type of test given the assumptions you have for SNP data. Of course, the assumption when using SNPs is that these are not causal mutations. However, we expect that in a largely dense SNP platform, SNPs are in linkage disequilibrium with the causative mutation (i.e. QTL or QTN).
That being said, we have the statistical assumption that variation on SNPs results in (i.e. is associated with) variation on the response variable. The opposite is not expected. You do not expect that non-responders will cause SNPs to change, as the central dogma of molecular genetics works the other way around.
Therefore, the use of Fisher's Exact Test is not what you are looking for. This test makes no assumption on the relationship that Response To Drug is a result of SNP variation. You should think of this test as a correlation between categorical variables - remember, correlations does not mean causation.
However, when you work with a logistic regression, you are explicitly (and correctly!!) saying that the variation in the response is statistically caused by variation in SNP genotypes. Again, we know biologically that SNP markers do not cause anything!
So, you should go with the logistic regression as this is actually answering what you are looking for.
Please let me know if you have additional questions. Thanks, Nick
Thanks a lot Nick VL Serão !
What i am thinking about is to differente between responder,nNonresonder and partialresponder. Before this i was using a binomial regression (responder, Nonresponder) but now i think i have to use the multinomial logistic regression. Atleast i get 2 rows with the results. What i am not sure about how i can interprete this. I am using SPSS and it is defining Partialresponse as the reference. Does it make sense?
Thank you for your help and advice.
Tristan Klintz ,
I do not use SPSS, but there are some things to do when using multinomial regression. First, my question is: are you fitting a regression line across genotypes (i.e., the allele substitution effect) or are you fitting categories of genotypes?
Depending on the way you fit SNPs (let's call them genotypes 0, 1, and 2), your results can be generated in two ways:
1) Obtain the expected proportion of Non/Partial/Responders for each genotype (0, 1, and 2), or expected probabilities of classifying as Non-Responders or Responders (compared to Partial-Responders) as the genotype increases (the allelic substitution effect).
2) Calculating the odds ratio of classifying them as Non-Responders or Responders (compared to Partial-Responders) for each genotype or as the genotype increases.
Although I am not familiar with SPSS, I believe the two approaches can be obtained from it.
Please let me know if this helps. Thanks, Nick
- Badges
- Science topic
- Similar topics
- Chemistry
- Pharmacology
- Pharmaceuticals
- Drugs