Interpreting P and q values in the results of genomic data analysis
Вставка
- Опубліковано 10 лип 2013
- If this was helpful this please give a "thumb up". Otherwise, leave a comment so I can improve the content - thanks!
This video has been made a primer to the use of P and q values that are used to indicate the significance of results in genomic data analysis and elsewhere.
It is based upon an article presented by the company Nonlinear on their web site:
www.nonlinear.com/support/prog...
[EDIT May 2015]
The above URL now redirects to:
www.totallab.com/products/same...
If you have any comments please let me know. - Наука та технологія
very well explained.
My understanding is that more tests give rise to a higher rate of false positives; i.e. the % risk increases with test number. So if two populations have the same mean and you do 100 tests, at p=0.05 you will actually have more than 5 tests that suggest that the means are different. Bonferroni corrects the p at which you accept a significant difference of means, given an input p and the number of tests. So assuming that Bonferroni is correct, microarrays have many more false positives than the t-test suggests.
Caroline Dahl Thanks for your interest in the video! I am by no means an expert and agree with you, however I thought I would clarify the last point. My interpretation is that although microarrays may have many more false positives than the T-test suggests, the P-value correction is designed to deal with that fact. .
are you sure about biological and technical replicate. please check it out ????
does increasing the sample size affect the FDR threshold??
+Ray ray Hi. I am no expert, however the FDR / adjusted value is an indication of the false discovery rate for the tests performed in the experiment. So whether there are 1000, or 10,000 experiments does not change the meaning of the qvalue. Ideally, I imagine, using the same value between experiments would be ideal, but generally you use a qvalue and log fold change to give you a manageable number of gene to analyse.
Correction
Slide titled: Multiple test P-value adjustment methods
Bonferroni correction method is P-value divided by number of tests performed (NOT P-value X number of tests as presenter claimed).
Example - Microarray gene expression data with 20, 000 genes test.
Bonferroni correction is 0.05/20, 000. Hence, adjusted P-value would be 0.0000025.
+Abdirahman Ali Thank you for spotting the mistake! I have added a note to the video.
That is a misleading way to phrase it. A Bonferroni-adjusted p-value is p * number of tests. What you are describing is a FWER-controlled threshold, which is indeed p-threshold/number of tests. However in the latter case the p-values have not been adjusted, the threshold has.