Using the MaxQuant algorithm, I am analyzing some Orbitrap data from a label-free LC-MS/MS experiment.
However, I encounter some difficulties in compiling the Experiment design template excel. Below is my design:
To compare protein differential expression:
- Experiment: 2 groups (control and treatment);
- Each group has 8 animals (biological replicates), and
- Each animal were run in triplicates (3 technical replicates).
Total = 48 Orbitrap result files (.RAW).
May you please shed some light on this? Thank you.
Hi Wong,
Technical replicates are not essential for the analysis when you have biological ones. In cases where biological replicates have not been done one can do technical replicates and use the same statistics, although the assumption of independence of samples is violated.
In the experimental design I would name each combination of animal and experiment (control or treatment) as a different experiment e.g. ControlA1 for animal 1 in the control experiment and TreatmentA2 for animal 2 in the treatment group and so on. I would name the technical replicates using the same experimental name e.g all three technical replicates for animal 1 in the control experiment would be ControlA1. This can be done because downstream statistics would be based on the biological replicates and you would sort of average out the results of the technical replicates.
With this setup you would get one value per protein per animal per condition. You need to take care to set proper group names if you would use the Perseus statistical framework and perform t-test statistics.
Ilian
Hi Wong,
Technical replicates are not essential for the analysis when you have biological ones. In cases where biological replicates have not been done one can do technical replicates and use the same statistics, although the assumption of independence of samples is violated.
In the experimental design I would name each combination of animal and experiment (control or treatment) as a different experiment e.g. ControlA1 for animal 1 in the control experiment and TreatmentA2 for animal 2 in the treatment group and so on. I would name the technical replicates using the same experimental name e.g all three technical replicates for animal 1 in the control experiment would be ControlA1. This can be done because downstream statistics would be based on the biological replicates and you would sort of average out the results of the technical replicates.
With this setup you would get one value per protein per animal per condition. You need to take care to set proper group names if you would use the Perseus statistical framework and perform t-test statistics.
Ilian
Also you should check the MaxQuant forum for similar questions/answers:
http://www.maxquant.org/
Link under "MaxQuant Help"
Ilian
Hi Wong,
In some of my proteomic projects I run technical replicates, in a similar manner to what you describe, to increase my number of quantifiable proteins and quantitation accuracy.
Prior to running the t-test, I then combine all the technical replicates sample-wise by taking the median, to avoid violating the assumption of sample independence in the t-test as stated by Ilian. In my experience, the biological variance between different samples has always been vastly larger than the technical variance.
Steps in the analysis: 1. Categorize all replicates together sample wise (e.g. "Control1" for all replicates of animal 1 and "Treated1" etc, as stated by Ilian). 2. Combine the replicate groups by taking the median (or mean). You can also require the protein to be quantifiable in at least two of your three technical replicates at this step. 3. Finally, categorize the samples as "Control" or "Treated" and perform the t-test of your choice.
Hope it helps and good luck with your research!
/Tue
Dear Wong,
I also work with technical replicates. But I don't like the idea to use them for statistics like ttest. Technical replicates will always show less variation than biological ones. I usually have the technical replicates as separated experiments in Maxquant. i.e. BR1_TR1; BR1_TR2. I don't know how Maxquant will treat them when yo use the same experiment name, so I recommend to use separate exp names. I asked the how Maxquant deals with this in the Maxquant googlegroup but I didn't get an answer.
For data analysis I calculate the mean LFQ of the technical replicates for each biological replicate. This might help to reduce technical variance. Afterwards I do the statistics just with the biological replicates (usually I do log2 transformation of LFQs before I run a ttest).
If you do statistics on all LFQ values of the technical replicates you will get a artificial increase in the N-number of experiments (24 replicates instead of 8 in your case). I personally don't like this. However, I am no statistician ;-)
- Badges
- Science method
- Similar topics
- Biological Science
- Genetics
- Gene Expression
- Protein Expression