If I remember rightly, eBayes is used to estimate the variance of expression for individual genes based on the expression profile of all the genes as a whole. I have a situation where I perform limma with eBayes and have repeated it using t.test and the multtest library (so without variance estimation) and have found that the p.values from limma are larger than those found using the traditional approach. Arising from this I have a few questions: 1) When should use the eBayes estimate and when should you not? 2) Is there anything wrong with using the results from using t.test and multtest? 3) Is there a way to use limma without using the eBayes correction? Many thanks Dan -- ************************************************************** Daniel Brewer, Ph.D. Institute of Cancer Research Email: daniel.brewer at icr.ac.uk ************************************************************** The Institute of Cancer Research: Royal Cancer Hospital, a charitable Company Limited by Guarantee, Registered in England under Company No. 534147 with its Registered Office at 123 Old Brompton Road, London SW7 3RP. This e-mail message is confidential and for use by the a...{{dropped:2}}

Daniel Brewer wrote: > If I remember rightly, eBayes is used to estimate the variance of > expression for individual genes based on the expression profile of all > the genes as a whole. I have a situation where I perform limma with > eBayes and have repeated it using t.test and the multtest library (so > without variance estimation) and have found that the p.values from limma > are larger than those found using the traditional approach. Some of the p-values or _all_ of the p-values? I would be surprised if all the p-values are larger - some should get smaller as well. > > Arising from this I have a few questions: > 1) When should use the eBayes estimate and when should you not? If you are using limma, IMO you should always use eBayes(). If you have very few samples, this will help. If you have many samples it won't hurt, and it is fast enough that the cost of doing so is minimal. > 2) Is there anything wrong with using the results from using t.test and > multtest? Depends on how many samples you have. If you have enough samples that you can get a reasonable number of combinations, then the empirical null distribution from multtest should give good results, and you can argue that the empirical null is better than assuming normality of your data and using the t-distribution as the null. However, at some point the central limit theorem will kick in and you don't have to assume anything anyway ;-D. So if you have lots of replication either e.g., rowttests() in Biobase or the multtest functions are great. However, if you don't have many replicates I think you can argue that limma is better. > 3) Is there a way to use limma without using the eBayes correction? Yes. IIRC, Gordon even mentions how to do it in the limma User's Guide, along with the advice that you shouldn't do it. Best, Jim > > Many thanks > > Dan > > -- James W. MacDonald, M.S. Biostatistician Affymetrix and cDNA Microarray Core University of Michigan Cancer Center 1500 E. Medical Center Drive 7410 CCGC Ann Arbor MI 48109 734-647-5623