I am using plink software to create an assoc file to get the P values and the number of SNPs which are in lakhs. The minimum P value that I am planning to use is 0.05 but since there are multiple tests I need to use the bonferroni correction. I think I should divide 0.05 by the number of SNPs, but if I do that the value I get is too small and none of the SNPs are below that range. One reason is my sample size, which is a lot less. Can anyone suggest how to select the significant SNPs from these?
To calculate the Bonferroni corrected p-value you multiply your p-values by the number of SNPs, any corrected p-values under 0.05 are significant. Bonferroni is considered to be v.conservative so you might also look at an alternative such as the Benjamini-Hochberg.
Do you use R/BioConductor at all? There is a BioConductor package called "QValue" that will readily convert any simple list of p-values into Q-values (aka an FDR corrected p-value). The manual will detail the settings needed to compute Benjamini & Hochberg (1995) FDR values, or use the revised methods of Storey et al.
Thank you so much., I will try and let you know if that solves the problem..
Colin Clark Sir, Before I start, I thought the P value should be divided by the number of SNPs to get the significant SNPs beacuse from what I read it says that p value should be divided by the number of tests in case if it is multiple.
Michael Black Sir, My major is Biotechnology and I am working on population genetics for which I am learning these and coming across these for the first time practically.. So, I am just a begginer and a learner to this field. If possible can u give me an answer that will make me understand quite easily..
I think the reason there's a disagreement on what to do is it depends whether you're referring to the p-value threshold for the test or the p-value for a particular SNP.
e.g. if you use a 5% significance threshold with 100 tests then you either
a) divide this threshold by 100 to make the Bonferroni-corrected threshold 5e-04
OR
b) multiply the p-value for each SNP by 100 and stick with 5% significance. So then a particular SNP with a p-value of 5e-03 would become 0.5 and no longer pass the threshold.
The two procedures are equivalent, as I hope you'll see. It should be obvious if you've done it correctly since it should always exclude more results as not significant.
The Bonferroni test is quite a stringent correction. Other measures such as FDR (suggested above) might give you different results and help you to find interesting cases. Remember a statistical test is only as good as the null model it assumes - a negative result doesn't necessarily mean there's no signal although you might well need to clean up the data a bit.
Thank you michael sir. That was simple enough to understand..
If you use Plink there is the command "--adjust". It will perform Bonferroni and other ways of correction.
Thank you sir. I did that and got the values..
hello all,
I too have a similar problem with my data wherein i have tried to associate 6 genotypes (all taken together as independent variable) with the different quantitative traits (which has been taken as the dependent variable, each individually) using linear regression in spss, now how can i correct the p values obtained ?
- Badges
- Science topic
- Similar topics
- Bioinformatics
- Bioinformatics and Computational Biology