Dear All,
I have a problem to do stat analysis to determine significant difference between control group and each condition group (vary dose of H2O2 treatment to cell line) of a target protein in western blot. The target protein is related to apoptosis pathway, so it should be increased in H2O2-treatment group. Yes It does, however the control group, which has no treatment, doesn't show the target protein band at all (I load about 20ug, not faint, just doesn't have it). I repeat experiment about N3-N4. All control group shows no band of the target protein (loading control shows up fine). When running stat analysis using one-way ANOVA to compare control group and H2O2-treatment group, it shows no significantly difference! due to qunantitative data from control group is too vary (may be due to background or whatever).
From what I see it should be significant difference, because no band VS show-up band is quite difference.
My question how should I do stat analysis in the case like this? not compare to the control group?
Thank you
Why do you need stats? If you have a good blot with good loading controls, why would anyone ask if you see a difference.
Hi Asa,
I would think presence of a new band in comparison to the target group would be enough to show the difference is significant. Especially when you are not quantifying the amount.
However, if you are set to do stat on your results, I would suggest to use “binary regression analysis” on your results. Binary regression analysis compares only two outcomes: present or absence (or in IT terms 0 or 1).
Good luck,
Ali
Hi Asa,
I would think presence of a new band in comparison to the ctrl group would be enough to show the difference is significant. Especially when you are not quantifying the amount.
However, if you are set to do stat on your results, I would suggest to use “binary regression analysis” on your results. Binary regression analysis compares only two outcomes: present or absence (or in IT terms 0 or 1).
Good luck,
Ali
You have to re-probe your membrane with a house keeping gene. Divide the OD (signal) of your gene by that of the house keeping gene and use for comparison. Or just use a sort of qualitative expression for your data like positive and negative. Also the statistical program can deal with the qualitative data and you can convert it to 0 and 1 values for example in the definition of the results in your program .
When there is no band in control and band in condition, this is clear enough that statistics are not needed.
as long as there's such big difference as you described , no band in control and new band in conditoned group , and the loading control (for example actin or GAPDH) show no difference , it's good enough
Helpful answers are already added in the post...BUT if you are determined to show the statistics then simply take the corresponding area when their is no band and divide its intensity with your condition group (where you have band)...simply it would give you the stats...do it in 3 replicate sets and showed it into graph...cya!
Roland Pochet was right. You don't need to show stat. data for yes and no. However, you can run an ELISA or In-Cell Western to show quantitative data in triplicate and that will give you statistically significance differences. Binary regression analysis determines significances usually used for disease state based on yes or no. Good luck.
If you need data in form of fold change increase, you need negative control band as well. Increase loading amount or sensitivity od your read out. If it would not help, increase concentration of your target protein via ultrafiltration. Of coure, the same treatment you have to do with H2O2 sample either.
This is a really tricky problem, and I have real doubts about the validity of data from scanning blots. In effect this type of data is the same as performing an ELISA with no standard curve and only one dilution. If it were me I would analyse one pair of samples (control and treated) and perform a whole series of dilutions and then generate a curve for each group for both loading control and target protein to determine the linear range for both groups for the target protein. I would then select concentrations that are within the linear range and select an authorad for the loading control wherein the bands are not saturated at for these concentrations and use these to normalise your target to loading control. I would the re-plot the curve for the normalised target data and measure the gradient for both groups and compare them. This is easier if you process your western blots by directly measuring such as LICOR or you use a phosphor imager.
Good luck!
Dear Andrew,
Would you mind providing a spreadsheet or any document to show us how do you quantify the western blot results? I think it will be better if you explain it by showing us some figure or data
You can try Licor In-Cell Western, you need to design experiments with time- and dose-dependent to show those variations. If performing only Western, repeat the experiment at least twice with 60micro gm of total protein loading per lane. Do densitometry and analyze it by Student t test for real values. I am sure that you will get statistically significant differences between controls and treated groups. Thanks, and good luck.
It seemed odd that your measure varies that much if there's no band. I'd imagine values are all near zero. Couldn't tell what your values look like. I assume there was a control gene and you normalized (divided) by it. If there were actual zeros in the data then taking logs is problematic, so I take log(normalized_value + C) where C is a constant that can be hard to choose (I pick it to sound good, and ideally "stabilize the variance" - the plot of stdev vs mean should have slope about zero). It sounds like your design is a two way anova ("repeated measures ANOVA") where 1 factor is the conditions and the other factor is the repeated experiments - but if that were actually true I think you would not be saying the experiment is repeated 3 or 4 times (my interpretation of what you are saying). It would either be exactly 3 or exactly 4.
My main points:
1) Adding constant before log-transform can help. It's common in old array data. Folks do it for count data from mRNA-seq too. Many other examples.
2) No adjustment based on the controls is needed - the experiment effects modeled in the ANOVA model. That's as it should be.
3) Analysis depends on the experimental design, which isn't that clear. The analysis is supposed to be planned before the experiment is performed.
To folks saying not to test hypotheses at all I am appalled:
I was trained by scientists, that if you do not do your best to carefully estimate the chances that the conclusions about your data are false positive, then it is not science. That's what quacks, pretenders, charlatans do. (There are times when decision theory is more important than hypothesis tests though. Then you are trying to maximize the subjective expected utility. Most biologists never think about that. Example: do I take the drug or not.)
- Badges
- Science method