Dear community,
Let me put in context my question:
I am working with qPCR to detect the absence/presence of a certain mutation in a given gene. However, the laboratorians that give me the data do not make serial dilutions on the input DNA samples, nor on the internal control. So the results they report, (the ct values), are based on the specifications of the manufacturer of the reagents.
Thus they have the following problem:
In some samples, the machine, (ABI 7500 FAST) reports a poor curve, and they are not sure if the CT value is correct. So what they do is, by visualizing the quality of the curve and the CT value, if the curve of the amplification is good, and the CT is above a threshold told by the manufacturer of the internal control positive gene, they report the result as positive or negative.
Given the fact that they do not perform serial solutions, even on the housekeeping genes, I cannot perform the ddct method. Mainly because this method is based on efficiency, furthermore, it is assumed that the efficiency is 100%. However, the efficiency, if there are serial solutions, can be computed and correct for the error.
Thus they do not perform serial solutions at all, and they ask me to make a method for objectively setting a more reliable threshold.
Is there a statistical/probabilistic based on data, so I can report a CT value in a more reliable way?
The data I have contains CT values of:
Sample gene of interest Sample Housekeeping gene ( I do not know which ) Positive control of the gene of interest Positive control of the housekeeping gene Negative control value of the gene of interest (it should be always zero or NAN but is not the case) Negative control value of the housekeeping gene (same as above)
I noticed, that the housekeeping positive control gene and the positive control gene of interest, have a statistically positive correlation.
A possible solution to my question, but I am not sure It could be a possible solution, to compute a ROC curve? \ Let's say, if I have 5 reactions, where each reaction is a mutation site, I have then 5 ct values for a given sample on different sites of the gene.\ Then, what I was thinking is that, if a sample is mutated in one region of the gene, it would be a CT value for that reaction but not on the other sites.|\ Thus, if I take the mean (with the removal of NA), for all reactions for a given sample, and take a sequence of CT values, ( I take the CT values as an independent variable), knowing the ground truth (given by the laboratorian), I can compute the specificity and sensitivity of the test for a set of CT values, then a ROC curve can be computed, the maximum of the ROC curve would be the "optimal" value, given a confidence interval..
What do you think?
Its a little unclear to me here what your aim is. You are talking about using qPCR to detect presence/abscence of given mutations - that is using qPCR for genotyping.
However, further down you talk about doing ddCt with housekeeping genes, which is usually what you would do if you were measuring gene expresison, but as far as I am aware ddCt is not used for genotyping.
Ok, so let's forget about ddct method. Of ddct if for gene expression measurement. And I only use it for genotypic, then based on some clinical data I could use. Logistic regression and perform ROC curve?