0

I am facing an issue, while trying to test for significance my data issue from mass spectrometry. I am working in the omics domain, where data collected have a natural variance that can reach 30%. To be clear, if I do my experience twice with the same exact sample, I will not get the same data, some intensities will vary up to 30%. This said, I would like to compare two experiments, considering this natural variance. I tried a lot of test(wilcox test, t-test, mood test, Kolmogorov-Smirnov test), for but as the amount of data I have is high, a little change in mean median or distribution, will make the test to reject the null hypothesis. enter image description here

As an example, experiment 5 (up and down) it is not ok as the boxplot are totaly different, however experiment 7 is, considering the natural variance.

Is there a test that would fit better to my scenario? Alternatively, I could try another approach.

Babas
  • 121
  • 3

1 Answers1

1

You could take two approaches:

1) If you are measuring each sample only once, then you can just run a t-test to compare your 2 experiments (or ANOVA for >2 experiments). But you will need a large sample size to accurately estimate/test for differences between experiments.

2) If you are measuring each sample multiple times, you can use a linear mixed-effects model with a random intercept for 'sample'. The random intercept will account for variance explainable by having multiple measurements on the same sample.

Take a look at the mixed-model tag, especially this useful question for more information. The wikipedia page on mixed-effects ANOVAs is also useful.

mkt
  • 11,770
  • 9
  • 51
  • 125