Multiple T-test

93 views
Skip to first unread message

niki...@gmail.com

unread,
Oct 19, 2017, 9:27:36 AM10/19/17
to ProteomicsQA
Hi there,

Graphpad offers multiple options to do multiple t-test corrections with the false discovery rate approach:

1. Two-stage step-up method of Benjamini, Krieger and Yekutieli
2. Original FDR of Benjamini and Hochberg
3. Corrected method of Benjamini and Yekutieli

Depending on which one I use, I have different significant proteins and with the first one I get more significant proteins that with the 2nd one (original FDR). Graphpad 7 has as default the two-stage step-up method of Benjamini, Krieger and Yekutieli. I am wondering whether to use this one  or to stick to the conservative originical FDR of Benjamini and Hochberg.

Thank you for your help.
Nikiana

Veit Schwämmle

unread,
Oct 23, 2017, 3:34:17 AM10/23/17
to niki...@gmail.com, ProteomicsQA
Dear Nikiana,

The most accepted and most commonly used methods are Benjamini Hochberg and the false discovery rates (q-values) by Storey (http://www.pnas.org/content/100/16/9440.full). The latter uses information from the actual distribution of p-values (as the Benjamini, Krieger and Yekutieli method) and therefore requires a certain amount of p-values (at least hundreds) distributed over the entire range (0,1).

It is well possible that you get more significant features this way as this method is less conservative. I suggest to compare the output of method 1 to the one from Storey. The output should be similar.

Best regards
Veit


--
You received this message because you are subscribed to the Google Groups "ProteomicsQA" group.
To unsubscribe from this group and stop receiving emails from it, send an email to proteomicsqa+unsubscribe@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/proteomicsqa/5873d2a6-88d6-4014-b145-423209307107%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.



--
            \|||/
           (o o)
 ----ooO-(_)-Ooo----

Don't worry about life; you're not going to survive it anyway.

niki...@gmail.com

unread,
Oct 26, 2017, 1:23:59 PM10/26/17
to ProteomicsQA
Dear Veit,

Thank you very much for your reply.

Best,
Nikiana


On Monday, 23 October 2017 08:34:17 UTC+1, Veit Schwämmle wrote:
Dear Nikiana,

The most accepted and most commonly used methods are Benjamini Hochberg and the false discovery rates (q-values) by Storey (http://www.pnas.org/content/100/16/9440.full). The latter uses information from the actual distribution of p-values (as the Benjamini, Krieger and Yekutieli method) and therefore requires a certain amount of p-values (at least hundreds) distributed over the entire range (0,1).

It is well possible that you get more significant features this way as this method is less conservative. I suggest to compare the output of method 1 to the one from Storey. The output should be similar.

Best regards
Veit

2017-10-19 15:27 GMT+02:00 <niki...@gmail.com>:
Hi there,

Graphpad offers multiple options to do multiple t-test corrections with the false discovery rate approach:

1. Two-stage step-up method of Benjamini, Krieger and Yekutieli
2. Original FDR of Benjamini and Hochberg
3. Corrected method of Benjamini and Yekutieli

Depending on which one I use, I have different significant proteins and with the first one I get more significant proteins that with the 2nd one (original FDR). Graphpad 7 has as default the two-stage step-up method of Benjamini, Krieger and Yekutieli. I am wondering whether to use this one  or to stick to the conservative originical FDR of Benjamini and Hochberg.

Thank you for your help.
Nikiana

--
You received this message because you are subscribed to the Google Groups "ProteomicsQA" group.
To unsubscribe from this group and stop receiving emails from it, send an email to proteomicsqa...@googlegroups.com.
Reply all
Reply to author
Forward
0 new messages