There are two main reasons for this problem: either one or more columns in the model matrix are linear combinations of other columns, or there are levels of factors or combinations of levels of multiple factors which are missing samples. We address these two problems below Case 1: In the above table two variables contain exactly the same information. The software cannot fit an effect for batch and condition, because they produce identical columns in the model matrix. This is also referred to as perfect confounding.

Solution:

the batch effect cannot be fit and must be removed from the model formula. There is just no way to tell apart the condition effects and the batch effects. The options are either to assume there is no batch effect (which we know is highly unlikely given the literature on batch effects in sequencing datasets) or to repeat the experiment and properly balance the conditions across batches. Case 2: In the above table the variables are not identical, but one variable can be formed by the combination of other factor levels. In the following example, the effect of batch 2 vs 1 cannot be fit because it is identical to a column in the model matrix which represents the condition C vs A effect.

Solution:

the batch effect cannot be fit and must be removed from the model formula. There is just no way to tell apart the condition effects and the batch effects. The options are either to assume there is no batch effect (which we know is highly unlikely given the literature on batch effects in sequencing datasets) or to repeat the experiment and properly balance the conditions across batches. Case 3: In the above case where we can in fact perform inference.In the above example the experiment comprises of grouped individuals, where the goal is to test the group-specific effect of a treatment, while controlling for individual effects.

Solution:

unfortunately the tool cannot be currently used for a similar scenario as above.We are working on it.You will have to use DESeq2 to define the design. Case 4: In the above case as highlighted a level is missing from a factor.The group 3 does not have level C

Solution:

Remove these samples with missing level.In the above case remove samples 13 to 16

Possible histogram versions and their meaning Case 1: If your p-value distribution looks like the above then You have (on the surface) a set of well-behaved p-values. That flat distribution along the bottom is all your null p-values, which are uniformly distributed between 0 and 1. Notice a rectangular shape with a peak at 0 .The taller the peak, the more p-values are close to 0 and therefore significant. Case 2: The above distribution is bimodal since we have two peaks.This u shaped histogram means low variance and therefore we need to correct for it.To correct for it please close this tab and select the checkboxes corresponding to the p-values to be corrected. Case 3: The problem with this distribution is that the N(0,1) null distribution of the Wald test was not appropriate for the dataset. In particular, the assumed variance of the null distribution was too high, and therefore we see a hill-shaped histogram of p-values distribution .To correct for it please close this tab and select the checkboxes corresponding to the p-values to be corrected. Case 4: The above is a flat distribution (what statisticians call a uniform distribution). At most a small percentage of hypotheses are non-null.No need to correct for such a distribution.