Can I get DE gene list for a simulated data set?

[tengfei-emory]

I am conducting simulations to evaluate DE gene detection performances. Is it possible to know which genes are differentially simulated among groups, so that I can have a gold standard to compare with? I appreciate your support! Thank you.

[lazappi]

For the Splat simulation you need to consider the DEFacGroupX columns in the row metadata. These tell you whether the expression of a particular gene has been changed in a particular group. If you have simulated two groups then a positive DE factor indicates that a gene is up-regulated in that group (and down-regulated for negative factors). It gets a bit more complicated if you have multiple groups. Other models return different information about intermediate parameters.

[tengfei-emory]

Thank you for answering! Could you please explain the complication when I simulate multiple groups? When comparing only two groups out of many groups, is it reasonable to ignore the DE factor for other groups and focus only on the two groups of interest? Your advice will be really helpful for our analyses. Thank you again.

[lazappi]

You are correct, you only need to compare the factors for the groups you are interested in. I will try to provide an example that explains how it works.

Let's say we are simulation some data with just two groups and we get these values for the genes:

Gene	Mean	DEFacGroup1	DEFacGroup2
GeneA	10	1	1
GeneB	100	1	5
GeneC	1000	10	5

In this case GeneA is not DE, GeneB is up-regulated in Group2 (foldchange = 5), GeneC is up-regulated in Group1 (foldchange = 2).

Now let's say we simulate three groups:

Gene	Mean	DEFacGroup1	DEFacGroup2	DEFacGroup3
GeneA	10	1	1	2
GeneB	100	1	5	1
GeneC	1000	10	5	8

We can do a similar thing to calculate the simulated differences in expression (foldchanges) between each of the pairs of groups we are interested in.

FirstGroup	SecondGroup	Gene	FirstDEFac	SecondDEFac	FoldChange
Group1	Group2	GeneA	1	1	1
Group1	Group2	GeneB	1	5	1 / 5
Group1	Group2	GeneC	10	5	2
Group1	Group3	GeneA	1	2	1 / 2
Group1	Group3	GeneB	1	1	1
Group1	Group3	GeneC	10	8	10 / 8
Group2	Group3	GeneA	1	2	1 / 2
Group2	Group3	GeneB	5	1	5
Group2	Group3	GeneC	5	8	5 / 8

The simulated foldchanges we have just calculated can then be used to tell us which genes are DE between two group (those where FoldChange != 1).

If we want to compare one group to multiple other groups it gets a bit more complicated again. We need to look at the DE factors in each of the groups but weighted according to the number of cells in each group.

Let's say we want to compare GeneA between Group1 vs (Group2 + Group3) we could calculated the simulated foldchange as something like:

FoldChange = DEFacGroup1 / ((DEFacGroup2 * nCellsGroup2  + DEFacGroup3 * nCellsGroup3) / (nCellsGroup2 + nCellsGroup3))

Hope that helps explain things! 😸

[tengfei-emory]

The explanation is really thorough and helpful. Thank you again!