Hello,
I need some advice about how to statistically analyze WB data.
I am trying to compare the levels of two proteins in mutants relative to wt samples (it's mouse embryonic tissue).
I repeated the blot 3 times, with control and mutant samples in each run following exactly the same protocol each time. (Because of the type of sample, it's not easy to increase the number of blots or to run all samples in the same blot, which would be better for analysis). Then, for each blot I obtained the relative intensities and normalize with Actin to get the normalized relative intensities (NRI) for the wild type and mutant samples.
Visually (and checking the NRI), the differences are very obvious but I would like to see if the differences are statistically significant.
So my main question is about the statistical analysis of these data. After looking for information and not finding much, what I have tried is to use Linear Mixed-Effects Models with genotype and protein (plus their interaction) as a fix effect, and run as a random effect, then do an ANOVA and calculate the pair-wise differences between groups. Run is not significant and the level of one of the protein is, which is consistent with what I see in the blots.
Is this analysis correct? Does anyone know any other proper way (if there is any) to analyze these data?
Thank you in advance
The NRI should have a right-skewed distribution. To use standard techniques I suggest analyzing log(NRI).
Your predictor is genotype (a factor with the levels "wildtype" and "tumour"). Thus, the difference in log(NRI) dpending on the genotype can be modelled by a very simple (standard) linear model log(NRI) ~ genotype (actually a simple two-sample t-test). However, not all log(NRI)'s are measured on the same blot, and there might be systematic differences between the blots. I would consider this using the blot as a random factor in the model: log(NRI) ~ genotype + (1|blot).
If you have several proteins, you may analyze them separately, or by a model including the genotype*protein interaction. As the blot-effect may be different for each protein, there should be a crossed or nested random effect for blot and protein (depending on your design, if the same protein is on different blots or not).
The NRI should have a right-skewed distribution. To use standard techniques I suggest analyzing log(NRI).
Your predictor is genotype (a factor with the levels "wildtype" and "tumour"). Thus, the difference in log(NRI) dpending on the genotype can be modelled by a very simple (standard) linear model log(NRI) ~ genotype (actually a simple two-sample t-test). However, not all log(NRI)'s are measured on the same blot, and there might be systematic differences between the blots. I would consider this using the blot as a random factor in the model: log(NRI) ~ genotype + (1|blot).
If you have several proteins, you may analyze them separately, or by a model including the genotype*protein interaction. As the blot-effect may be different for each protein, there should be a crossed or nested random effect for blot and protein (depending on your design, if the same protein is on different blots or not).
(More than 8 months later) thank you for your advice, Jochen Wilhelm
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis