Question: The Most Appropriate Statistical Test
0
gravatar for cfarmeri
4 months ago by
cfarmeri130
Japan
cfarmeri130 wrote:

How to choose statistical test method appropriate to my experiment design?

I have following RNAseq samples. (I know 2 replicates is not enough...)

  1. WT : wild type (replicate1, replicate2)
  2. mm : gene X mutant (replicate1, replicate2)
  3. OE : gene X overexpression (replicate1, replicate2)

To identify DEG(differential expression gene), I prepare these hypothesis.

・Null hypothesis : there is no expression change when manipulating gene X expression.

・Alternative hypothesis : there is some expression change when manipulating gene X expression.

I think ANOVA-like test in edgeR is appropriate method so far. In using this, multiple test correction is unnecessary and comparison between samples more than 3 is available. (In edgeR user's guide, ANOVA-like test is 3.2.6.)

However, I have not ever seen these experimental RNAseq design, so cannot choose the most appropriate statistical method. So I'm wondering if anyone knows some advice about this problem...

ADD COMMENTlink modified 4 months ago by h.mon15k • written 4 months ago by cfarmeri130
0
gravatar for Hussain Ather
4 months ago by
Hussain Ather860
National Institutes of Health, Bethesda, MD
Hussain Ather860 wrote:

ANOVA should be fine for testing contrasts between conditions.

ADD COMMENTlink written 4 months ago by Hussain Ather860
0
gravatar for h.mon
4 months ago by
h.mon15k
Brazil
h.mon15k wrote:

If you use the ANOVA-like test, you won't know where the differences are - they may well be counter-intuitive to your understanding of the underlying biology. Testing each contrast separately should give more insight - but keep in mind with two biological replicates, you have very low statistical power.

In using this, multiple test correction is unnecessary and comparison between samples more than 3 is available.

You mean, multiple testing due to multiple contrasts, but multiple testing correction due to multiple genes is still needed.

ADD COMMENTlink written 4 months ago by h.mon15k

I must disagree with you. In the case of having two classes comparison, t.test and anova is practically identical. However, you should use ANOVA whenever you have more than two levels (classes). You get better p-value. It is NOT just a trick or something, it make sense to use all samples (all information) instead of the samples in the two classes you want to compare, it gives better estimation of variance!

ADD REPLYlink modified 4 months ago • written 4 months ago by arta480

The ANOVA-like test he is referring is this:

An ANOVA-like test for any differences

It might be of interest to find genes that are DE between any of the groups, without specifying before-hand which groups might be different.

[...]

Technically, this procedure tests whether either of the contrasts B-A or C-A are non-zero. Since at least one of these must be non-zero when differences exist, the test will detect any differences.

If he tests for each contrast separately, he will know which treatment differs. If he uses the referred ANOVA-like test for any differences, he won't.

ADD REPLYlink written 4 months ago by h.mon15k
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: 633 users visited in the last hour