Hi,

1 - I am a bit confused by your presentation of your "pre-treatment". You talk about mean at conc1 and time 0 (or time 72h) - does that mean that you have more than a measurement at each time ? If so using mean(s) for a given occasion (time x conc) is not a good idea, you just mask variations that might be informative. The use of really observed data is advisable, even if it is not so handy to tackle them with the _drc_ package (a good one indeed).

2 - the "standardization(s)" you performed produce(s) more harm than good, as you mention with your "negative values".

3 - My guess is that you handle counts (number of confluence(s) may be ?). If so then you should consider using glm-models instead. If you want help let some original counts data available, we (I mean the ResearchGate community) will give you a hand if you give us the good data.

4 - The curves you display are so different that I can not believe there is not any significant difference between them.

1 - Let me try it again: I've got two cell lines (gr, ctrl) which were treated with a drug in different concentrations (0, 0.001 etc.), each concentration was done in technical quadruplicates. I've meassured the confluence right after treatment (0h) and after 72h.

2 - Any suggestions for a better normalization?

3 - Yeah, I've measured cell confluency using IncuCyte

4 - should I just present the SD of the replicates for each concentration?

Thank you very much!

Hi Oli ,

please find attached a reshape of your data I ask you

1 - to check carefully the matching of values for 0h and 72h;

2 - to tell me if the quadruplicate :

- is made on the same biological material (well, plate, colony, culture or whatever) by a "physico-chemical" reading repeated four times (in such a case the column "object" of the file should be labelled "G1" repeated four times, "G2" repeated four times, etc;

- or if the biological material is "different" for each reading, in such a case the objects are to be labelled "G1" to "G36" (the actual choice with no commitment at all) and "C1" to "C36" for the ctrl group.

This point is very important for an apprpriate statistical treatment.

I will wait your response before any descriptive treatment, not to mention model fitting with or without standardization.

All the best

Dear Jean

1 - the data seems to be correct.

2 - quadruplicates are based on the same culture, but are in different wells (96 well plate as you can see in the data I gave you)

Best,

O

Dear Oli,

my final cut attached - I added two columns with codes for the cell cultures and for the repetitions/wells (might be superfluous but one never knows) - tell me if that makes sense. If you agree, I propose to explore a bit your data with some graphs so we could elaborate a strategy for an analysis.

Best.

Good Morning,

looks good to me.

What would you do differently?

Hi Oli,

first I tried to build a large picture of your data. That means that I am trying to put on the same graph, in an intelligible way if possible, all or almost all the information available.

Namely the observed values at 0h and 72h, their differences, with the dose range, differentiating between cell.lines. I produced the attached picture.

On this picture you have each line cell (Gr="gr" and Ct="ctrl"), green (0h) and black (72h) for "ctrl", blue(0h) and red (72h) for "gr), displayed by (log)dose (with an arbitrary shift for dose 0), labelled 0 to 9. The arrows are the evolution between time 0h and 72h for each cell line.

If you consider it carefully you may/might agree that 1) the differences for "gr" are quite constant and , in my opinion, there is no clear need to standardize further than a mere difference between 0h and 72h for this cell line.

For "ctrl" the evolution is different, you have two _plateaux_ for the differences, one below dose #3 (positive differences), one above dose #4 (negative or nearly null differences), with a sharp decline between doses #3 and #4/5.

My opinion is that the differences are a fair description of what is seen on your data. Before to start a statistical analysis or any model building, it would be wise to write down what the question(s) is/are.

Hi Jean,

wow - thank you for putting so much work in to it.

There are two questions:

1 - What is the EC50 for each cell line?

2 - Is there a significant difference between them?

Main idea was to plot a dose-response-curve on it. The idea behind my "normalization" was to show a %-difference of control (being the 0 dose).

Hi Oli,

The EC50 might be defined for the cell line "ctrl", between dose #3 and #5 certainly, but it seems hard to assess an EC50 for the cell line "gr" where the differences (apparently) do not change with the doses, hence the fitting of a dose response curve could be problematic with this cell line ("gr"), hence the definition of an EC50.

That stated, there is a huge difference between the two cell lines, the differences are much higher and constant through doses with the cell line "gr". For this point it could be possible to treat the question by anova, circumventing the definition of a dose response model.

Best.

Hi Oli,

after some thought I reached an idea. If your question is the comparison of EC50, I think that you may have to reconsider the doses you used.

The dose-response curves are usually S-shaped (sigmoids) increasing or decreasing (as in your case). To estimate properly the parameters of such curves, and consequently the EC50, you have to get enough observations on the _plateaux_ at both ends of the curve, but also some points (at least three or four in my experience) in the varying zone between the _plateaux_. You missed this point(s) with your doses.

For the "gr" cell line you get points only on the left _plateau_, for the "ctrl" cell line you have only one (or may be two) point(s) in the varying zone.

My advice would be to increase the doses for the "gr" line -- how much is an open question -- and to add some doses between actual doses #3 and #5/6.

For EC50 comparisons there is no technical need that the doses have to be the same for different groups.

Thank your very much. Right now there's is no possibility to redo this experiment. I Just want to find a way to present my obtained a better way. I know the data is not perfect in any ways.

I will try to elaborate a (statistical) comparison of the two cell lines avoiding a parametric dose-response model unreachable, in my opinion, with the actual data.

I have some simple ideas to check, I will tell you soon if it works.

Bye.

Best.

Hi Oli,

you will find attached some results of my investigations of your case. Obviously too verbose, but I tried to clearly separate main points and secondary thoughts placed in appendices.

I hope it may help, you have plenty of R code you may reuse and modify at your convenience.

Best regards