Hi, am a final year student doing research on biomarkers for breast cancer, and I used microarrays to find my target genes. After identifying my target genes, I wanted to validate the microarray results using RT-qPCR. Had a few questions regarding the RT-qPCR process.
1. I did duplicates for each patient sample during my RT-qPCR, and rejected any samples where duplicates had Ct values with a difference of more than 1. I then further tried to identify any outliers by constructing a box and whisker plot and rejecting values above or below these thresholds. My question is, when constructing the box and whisker plot, do I construct a box and whisker plot for each data set (breast cancer, patients with family history of breast cancer, normal patients) or do I combine all the data sets together?
I would be careful with removing "outliers". If there are many "outliers" then your method or your analysis is flawed or inappropriate and this severely questions the whole biomarker, and if such "outliers" are rare they won't impact your analysis anyway. Removing "outliers" just because the data looks like usually produces severly biased estimates of the variability. But this is critical to judge if a biomarker is useful at all.
But if you look for "outliers", then look for them separately in each group.
And note that Tukey did not invent the boxplot to identify points that should be removed. The intention is rather to identify points where one might have a closer look if there is any known reason why these values are so different from the rest, and also to get a hint if the normal probability model is appropriate to describe the variability of the data.
Do not make your data fit to the analysis - make the analysis fit to the data!
Note that the distribution of dCt values should be approximately normal, but not the distribution of 2^dCt (sometimes referred to as "relative expression values") or some other non-linear transformations of the dCt values.
Remark about pcr
your sentense : I did duplicates for each patient sample during my RT-qPCR, and rejected any samples where duplicates had Ct values with a difference of more than 1
You should rerun your samples if you find this kind of duplicates,a criterum of 1Ct for rejection is much to high, 0,5 Ct is already high, so check your method first before use the stats.
,
Hi Jochen and Vincent,
Thanks for your answers. Am a bit of a novice in RT-qPCR as I haven't done it before throughout my undergraduate years until this project for my graduation thesis. Consulted my professor regarding the Ct values difference of more than 1 and he told me the maximum allowed is 1 Ct, I preferred a Ct value difference between duplicates of less than 0.5. However, by doing this, some of the standards required for the standard curve that I used for my experiment did not pass the grade, causing a bit of a dilemma. Due to time limitations as well as resources, I am not able to repeat the experiment. Would excluding values of more than 1 Ct + excluding box and whisker plot outliers in each group (breast cancer, patients with family history of breast cancer, normal patients) be a good enough quality check for my data?
- Badges
- Science method