Question: Save experiment with batch effects - Protein microarray
1
gravatar for s.lima.diogenes
10 months ago by
s.lima.diogenes10 wrote:

The situation is as follows

Collaborators used a protein array to detect changes in protein abundance between two experimental groups.

The simplified original design was something like this:

sample  /// expGroup /// subject /// timePoint /// batch

s1    ///  group1  ///  sub1   ///     0     ///   1

s2    ///  group1  ///  sub2   ///     0     ///   1

s3    ///  group2  ///  sub3   ///     0     ///   1

s4    ///  group2  ///  sub4   ///     0     ///   1

(...)

They wanted to find differentially expressed proteins (DEPs) between both groups in the same timepoint. Unfortunately, they did not find any DEP that passed a p-value threshold, but then again, groups were not that different. After that, they decided to test if they could find DEPs after a treatment. The experimental design then became like this:

sample /// expGroup /// subject /// timePoint /// batch

  s1   ///  group1  ///  sub1   ///     0     ///   a

  s2   ///  group1  ///  sub2   ///     0     ///   a

  s3   ///  group2  ///  sub3   ///     0     ///   a

  s4   ///  group2  ///  sub4   ///     0     ///   a

(...)

  s5   ///  group1  ///  sub1   ///     5     ///   b

  s6   ///  group1  ///  sub2   ///     5     ///   b

  s7   ///  group2  ///  sub3   ///     5     ///   b

  s8   ///  group2  ///  sub4   ///     5     ///   b

(...)

As it is clear from the design, samples from timePoint 5 were all assigned to batch "b" while samples from timePoint 0 were assigned to batch "a". And, as a matter of fact, clear batch effects were detected (different fluorescence distributions, PCA separation...)

I feel confident in performing these comparisons:

group2 - group1 in timePoint 0

group2 - group1 in timePoint 5

But, I don't feel confortable in testing this comparison:

5 - 0 pairing by subjects regardless of expGroup.

Anyway, I performed a DEP analysis (using paired limma and paired T test), and got meaningful results (Proteins expected to be DEPs had the highest fold-changes). Nevertheless, I cannot quantify the extent of batch effects in these results, so, I don't trust the p-values.

My questions are: Is there any test using subject fold-changes that are more robust to these spurious effects. I though of using rank-based methods, but I do not actually know if the results will be less affected by batch effects. Has anyone faced similar problems? What did you do? Is there any possibility of saving this experiment?

Everything was done before I had any contact with them and I cannot suggest to start everything all over

Thanks a lot!!

ADD COMMENTlink modified 10 months ago by Kevin Blighe66k • written 10 months ago by s.lima.diogenes10

What did you do?

I performed the comparisons that could be performed, and stated the other comparisons couldn't be made because they were confounded with batch effects.

Unless there is some control spike-in, I don't think there is much you can do.

ADD REPLYlink written 10 months ago by h.mon31k
Please log in to add an answer.

Help
Access

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.3.0
Traffic: 1236 users visited in the last hour