Hello all,
I've posted the below text on the rMATS user group already, but I thought I'd also try it here.
Many thanks for creating such a useful tool! My lab makes quite heavy use of rMATS, but I've recently encountered a confusing piece of behavior regarding the calculation of p-values that I would love to get some clarification on.
I have an event with an IncLevelDifference of -0.192, with 6 replicates in each sample group. The distribution of samples looks distinct when plotted, but rMATS reports a p-value of exactly 1. I understand it is possible to have a large effect size without statistical significance, but for this distribution to have a p-value of 1 seems rather mysterious.
I have uploaded a snippet(Groups won't let me attach it, so this is a Google Drive link) of the raw SE.MATS.JCEC.txt file, with the event in question (47518) alongside two other events from the same gene, which have more reasonable-looking p-values. I ran a Fisher's exact test as a quick sanity check on these events -- although this is of course not the exact calculation rMATS uses, the resulting value of p = 0.0002 for this event furthers my suspicion that the p = 1 finding is due to unusual behavior in the algorithm that calculates p-values. I have added this as a column to my minimal JCEC file for these three events.
rMATS was run under default parameters with paired-end sequencing on mouse data. I attempted to run with the parameter --cstat 0 after reading this GitHub thread indicating that this allows for more significant events, but still got p = 1 for this event. Any insight into why this is occurring or suggested alternative parameters would be greatly appreciated!
Many thanks,
Ben Carr
Hello all,
I've posted the below text on the rMATS user group already, but I thought I'd also try it here.
Many thanks for creating such a useful tool! My lab makes quite heavy use of rMATS, but I've recently encountered a confusing piece of behavior regarding the calculation of p-values that I would love to get some clarification on.
I have an event with an IncLevelDifference of -0.192, with 6 replicates in each sample group. The distribution of samples looks distinct when plotted, but rMATS reports a p-value of exactly 1. I understand it is possible to have a large effect size without statistical significance, but for this distribution to have a p-value of 1 seems rather mysterious.
I have uploaded a snippet(Groups won't let me attach it, so this is a Google Drive link) of the raw SE.MATS.JCEC.txt file, with the event in question (47518) alongside two other events from the same gene, which have more reasonable-looking p-values. I ran a Fisher's exact test as a quick sanity check on these events -- although this is of course not the exact calculation rMATS uses, the resulting value of p = 0.0002 for this event furthers my suspicion that the p = 1 finding is due to unusual behavior in the algorithm that calculates p-values. I have added this as a column to my minimal JCEC file for these three events.
rMATS was run under default parameters with paired-end sequencing on mouse data. I attempted to run with the parameter --cstat 0 after reading this GitHub thread indicating that this allows for more significant events, but still got p = 1 for this event. Any insight into why this is occurring or suggested alternative parameters would be greatly appreciated!
Many thanks,
Ben Carr