Dear King

Why you are subtracting is not clear. Moreover, you may think to use square root transformation for WT data. You may also think to calculate percent data. Pl check, whether it suits to you. 

Hi Rajiv,

I substract the TS-WT to look for the average differences between the TS and WT groups. The values were in log format and the differences between the group will indicate how many fold the TS group is higher or lower than WT.

What did you mean by calculate percent data?

Thank you!

I have no Idea what your data is, so it is difficult to give any specific help.

It looks as if the simple subtraction of the mean negative control values per plate is not very sensible for that kind of data. It might be more appropriate to normalize the data not by subtraction but by division instead. Another option is to use a normexp-model (a convolution of a model for normal-distributed "noise" and a model for exponentially-distributed "signal", that will generate strictly positive corrected values).

So your problem seems to be in the background correction. Here you should invest some research. Having fixed that problem will likely solve your current problem.

Hi Jochen,

Tqvm for your reply. In the experiment, we have 3 samples which were blanks. When we put our cells into it, we expect no colour changes (absorbance as measurable) would happen to these blanks. However, in cells, they have reserved energy when metabolised will lead to basal colour changes, hence the absorbance reading (raw values).

When we tested the cells on different substrates, certain substrates inhibit the utilization of reserved energy, therefore the the absorbance readings for these samples treated with these subtrates turned out to be lower than the blanks. In normal practice, we subtract the readings of the treated cells with the blanks as a measure of normalisation. Therefore, some readings were negative.

We have two type of cells here... the WT and TS. We repeated it 3 times each for each substrate. Therefore in the worksheet 8, the log-transformed values were tabulate for WT and TS for analysis.

Hope it makes sense to you now. Thank you for looking into it!

Michael

Then the untreated control cells do not give you a background signal for the measurement, and therefore you should not subtract the signals from these controls. Instead, you should use controls where the energy utilization is completely inhibited. This youd give you a technical background signal that may be subtracted (even then, a normexp model for nbackground correction would still be better).

As a second-best solution, you might consider testing the mean difference in the subtracted values (no transformation). The distribution of the residuals is pretty well normal (at least for the WT data; I did not check the TS data), so this may serve your  purpose.

Hi Jochen,

I really appreciate your time spent on the question. I didn't mention that the experiment is also carried out in a time course manner which I didnt include in the spreadsheet to reduce it complexity. If I used the sample that give the lowest background/reading for substraction (normalisation) at one timepoint, the same sample may not be the lowest in reading in a different timepoint. Would you suggest that I choose different sample in each timepoints for normalisation as long as the sample has the lowest reading?

Cheers~

Michael

No.

The control group should always be the same. If there are several groups that can serve equally well as control, then you can consider using all of them. I would again advice using a normexp model to correct the background...