Question

TREAT function - how to interpret the result

0

Entering edit mode

Sanches • 0

@sanches-13853

Last seen 6.7 years ago

Hello everyone,

I am so sorry for asking help for a basic issue.

So, I have used TREAT function to test for the differential expression of genes. For example, is it correct to use logFC to compare the relative expression between ADAM8 and SLC25A27? As you can see below, ADAM8 is the ranked but it has smaller logFC. So, in this case, what could do?

Thanks in advance

ENTREZID	SYMBOL	UNIPROT	logFC	AveExpr	t	P.Value
101	ADAM8	P78325	4.145521092163	8.67707140673464	26.5515520677772	6.92389718357961e-09
587	BCAT2	O15382	4.53846888171869	8.92879466212405	24.8502028681871	1.13492312522918e-08
6591	SNAI2	O43623	4.25097730022446	8.81891453855237	23.1172960098245	1.90935989746792e-08
9481	SLC25A27	B4DHR4	5.60733023152463	11.2411231360327	23.0872979003407	2.00545074568876e-08
27240	SIT1	Q9Y3P8	3.60236159436291	8.06007485742637	17.2725187626978	1.56307173434009e-07
7965	AIMP2	A0A024QZY1	3.76550433124359	9.60469812523139	17.2108174737464	1.60920592666298e-07

limma treat • 1.3k views

ADD COMMENT • link updated 7.1 years ago by Gordon Smyth 51k • written 7.1 years ago by Sanches • 0

score 1 · Answer 1 · 2017-09-18

Please do not reorder the genes by fold change. The genes have already been ordered by likely biological importance according to the treat approach.

Fold change is not the only consideration when judging significance or importance. One also has to take into account the consistency of the results across replicates. In your table, SLC25A27 has a higher fold change than ADAM8 but it is also shows more variability between replicates than ADAM8 and hence ends up with a slightly lower t-statistic. ADAM8 has a slightly lower fold change but shows clearer separation of expression between the groups.

The whole purpose of limma is to combine fold change and variability into a combination ranking of statistical significance and potential biological importance. If you reorder the results by fold change then you prevent limma from doing its job. I think that ignoring variability and just using fold change alone is unwise.