Q&A about MAGECK from MAGECK google group

When to use RRA or MLE in MAGECK? 

The short answers is, if you compare with 2 conditions, use RRA. If you want to compare multiple conditions, use MLE. RRA generally give you better estimation of p values, but MLE is really powerful in terms of multiple condition comparisons.

Reference:

https://groups.google.com/forum/#!msg/mageck/91Jm6vOPFi8/w9TeuJ2-sVYJ

How does MAGECK work? 

MAGeCK-VISPR usually calculates beta scores with 0-day as baseline. In other words, if you have 0-day, drug treated, and control samples, you will get two beta scores in mageck-vispr, one with drug treated samples, the other with control samples. MAGeCKFlute further tells you the difference between (drug-treated beta scores) vs (control beta scores). 

You can model the difference directly in MAGeCK-VISPR as well, by setting up the corresponding design matrix. However that is not a common approach and sometimes it's hard to interpret. MAGeCKFlute further gives you a more straightforward analysis on the gene behavior in two conditions, and the differences between two conditions.



What FDR cutoff to use? 

Hi Christian,

First, you should look at the FDR value, not p value. Usually FDR<0.25 can be considered significant; in this case it's only negatively selected.

Second, it's possible to have a gene that shows both pos/neg selected, especially if two sets of guides in the same gene show opposite direction.

Reference:

https://groups.google.com/forum/#!topic/mageck/2r96vmJsHws

Whether the number of sgRNAs per gene matters? 

Yes, the number of sgRNAs per gene matters, and in theory a gene with more sgRNAs will have smaller RRA scores. To control this, MAGeCK calculates permutation p values separately for genes with different sgRNAs, so remember to always look at their p values.

Reference: https://groups.google.com/forum/#!topic/mageck/MDJ8GFRnKm8