What is your opinion on batch correction in quantitative proteomics data sets.
When do you use it? What is the reasoning/measure. How do you apply it to your data? How do you set up your experiments to be able to use batch correction?
Thank you for sharing your expertise.
Batch effect correction is important. Separating batch effect from random error can gain the sensitivity on the test of differential abundance.
Although I prefer doing batch correction every time, batch correction makes more difference when batch effect is significant. If you have enough "house-keeping" or "background" markers, then batch effect could be seen and hence be corrected.
Nonetheless, randomization of experimental design is crucial, since you would like to make sure what you correct is batch effect instead of some effects caused by other latent variable such as wash-out period, how long the sample has been frozen, age or gender of patients, ..., etc.
Batch effect correction is important. Separating batch effect from random error can gain the sensitivity on the test of differential abundance.
Although I prefer doing batch correction every time, batch correction makes more difference when batch effect is significant. If you have enough "house-keeping" or "background" markers, then batch effect could be seen and hence be corrected.
Nonetheless, randomization of experimental design is crucial, since you would like to make sure what you correct is batch effect instead of some effects caused by other latent variable such as wash-out period, how long the sample has been frozen, age or gender of patients, ..., etc.
Hi Roland Bruderer,
Like Yin-Yang Cheng, I randomize my proteomics samples as early as possible in the sample preparation process to minimize time-dependent effects. However, in one of my research projects with many samples, an LC column replacement caused a strong batch effect in the data, which was easily visible on a principle component analysis. Unfortunately, it was not possible to rerun the samples, so for this project I used the ComBat algorithm with R, which turned out to be very effective. I was very impressed with the results which retained the biological changes in the data, while removing the batch effect. The algorithm has been cited in a number of peer-reviewed scientific papers, and details can be found from the reference below:
Johnson, WE, Rabinovic, A, and Li, C (2007). Adjusting batch effects in microarray expression data using Empirical Bayers methods. Biostatistics 8.
I hope this answers your question, and good luck with your research.
Sincerely yours,
Tue Bjerg Bennike, PhD
Hi Lin-Yang and Tue Bjerg,
Thank you very much for your input. It is indeed important and optimal to randomize sample preparation and acquisition. Thanks for the tip with ComBat.
We now use a protein wise batch mean adjustment, as default procedure. This is very simple and robust, if you have many samples in the batches. The variance is not adjusted and on Bayesian approach is performed.
- Badges
- Science topic
- Similar topics
- Chemistry
- General Biochemistry
- Proteomics