My experiment have several treatments (e.g. A, B, and C).
If I have interest in differentially expressed genes (DEGs) over all treatments, ANOVA will be good enough.
However, my major focus is identification of DEGs between two treatments, but came into trouble with adjusting p-values. What will be the best solution for adjusting p-values?
1. just perform multiple test correction (e.g. p.adjust function of R) using p-values from each pairwise comaprisons (e.g. A vs B)?
2. collect all p-values from all pairwise comparisons (A vs B, A vs C, B vs C) and run single mutliple testing correction and report q-values?
3. any best solution for this?
Thank you in advance.
ANOVA compares (nested) models; it does not tell you anything about the differences between individual treatments.
A difference is defined between two values, and the data is tested under the hypothesis of zero mean difference by a t-test. To test the differences between all pairs of groups you have to do a t-test between each possible pairwise combination of groups, ideally using a pooled estimate for the standard error, and possibly a correction for multiple testing.
In R you would simply use the function pairwise.t.test() to get the p-values. By default, it uses the pooled standard error and performs a Holm correction for multiple testing (but than can be changed, if desired).
ANOVA compares (nested) models; it does not tell you anything about the differences between individual treatments.
A difference is defined between two values, and the data is tested under the hypothesis of zero mean difference by a t-test. To test the differences between all pairs of groups you have to do a t-test between each possible pairwise combination of groups, ideally using a pooled estimate for the standard error, and possibly a correction for multiple testing.
In R you would simply use the function pairwise.t.test() to get the p-values. By default, it uses the pooled standard error and performs a Holm correction for multiple testing (but than can be changed, if desired).
Dear Jochen,
Thank you for the quick reply.
In my case, I used other test (several nonparametric tests) to identify DEGs between all pairwise combination of treatments (not certain if pooled estimate for the standard error is considered for this).
If I understood your answer, next thing will be just collecting p-values from all pairwise combinations and perform multiple testingin correction?
For example, if I have 1,000 genes, and 3 treatments, there will be 3000 p-values. I would run p.adjust function with 3000 p-values. Is this correct?
Thank you~!
Hi,
Assuming the data is normally distributed, I would opt for an ANOVA followed by a Tukey's posthoc test. Tukey's posthoc test is a single-step multiple comparison procedure and statistical test. It can be used on raw data or in conjunction with an ANOVA (post-hoc analysis) to find means that are significantly different from each other. It is essentially a t-test, except that it corrects for family-wise error rate, which accrues when there are multiple comparisons being made.
Hope this helps!
All the best!
Qin Xiang
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA