Hi
I have performed a gene card with 96 genes to see whether certain genes are expressed or not between skin models and full thickness skin. In some cases, genes are not expressed and so I get an undetermined value - how do I treat these when normalising the data via delta CT?
Many thanks
You cannot calculate a dCt where one of the Ct values is not defined. You simply don't have quantitative data in these cases.
One possibility to analyze such cases is to fall back to a qualitative analysis (analyse the proportion of negative and positive PCRs), another option is using a censored regression model (having "no value" is actually a "right censoring" at the upper limit of quantification), see
Neyman J, Scott E (1948). \Consistent Estimates Based on Partially Consistent Observations." Econometrica, 16, 1-32.
Tobin J (1958). \Estimation of relationship for limited dependent variables." Econometrica, 26, 24-36,
and the R package censReg in particular (https://cran.r-project.org/web/packages/censReg/vignettes/censReg.pdf).
Both approaches will likely require relatively large sample sizes to useful precisions. Censored regression models may not give results when all the values in one group are censored.
Thanks for all your replies - would it be suggested to set all values to 40 and then use a filtering system to exclude gene if 70% biological repeats (i.e. full thickness epidermis) are greater than 37ct?
Also, at which point do I combine my repeats - I have 3 full thickness epidermis and 3 model epidermis from different patients (with 3 separately built models for each patient) - are these separate models biological or technical replicates and how do I deal with them?
Many thanks once again - I'm new to all of this and really want to get going on it
:)
I dont see the point in setting undefined Ct to 40 to filter. Why not directly filter genes that do have undefined Ct values?
The proplem remains for genes that will have some (less that 70% or whatever) undefined Ct values. If you use only the PCRs that do have a defined Ct you will systematically overestimate the expression of this gene (-> bias). It is ok as long as you consider the maximal bias in the judgement of differential expression.
I do not understand well what your research question is, but it looks that the patient is the "biological replicate", the epidermis models are your (treatment) groups. As you have cell cultures you can have (tecnical) replicates at different levels (isolation from the same patient, culture plate/well, RNA extraction from the same plate/well, and the PCR itself. If one wants one can thoroughly the contributions of all these replications on the variance of your data using hierarchical (mixed/multilevel) models, but that is useful only if one really need this information (e.g. to better plan future experiments). Biologically relevant is only the biological replication, and you will get the correct statistics for the biological variance in your assay after averaging all technical replicates for each biological replicate (patient), so that in your case the analysis will be based on 3 values per group.
Thanks for your reply Jochen
I'm really struggling to work out how to analyse the data with these undetermined values - largely they coincide with other samples having high Ct values, so largely I would guess that the genes are unexpressed but I'm really unsure of how to address them. I've found this paper (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4133581/) but I'm not familiar with R and therefore cannot interpret the data. I could change to 40 but like you said, this will introduce bias and I only have single technical repeats and so cannot exclude so I'm quite stuck on what approach to take
I thought about doing grubbs test for outliers also but I'm unsure at which stage to do this - with Ct or with delta ct values and at which point to combine the repeats of the models
- Badges
- Science method