DE Genes by Perturbation

[Eduardo Gade Gusmão]

Dear sir/madam,

I couldn't find this information in the paper or on the supplement material. So I have a couple of questions:

1. Do you have, by any chance, the DE expressed genes by perturbation?

2. What is exactly the mm3.xlsx supplement table? The regulatory coefficients?

3. Can I extract the "important" genes in each perturbation from the mm3.xlsx table? If so, what would be a good cutoff threshold?

Thank you very much!

[Atray Dixit]

Hi Eduardo,

Yes you can to some extent extract the important genes in each perturbation from the mm3.xlsx table. If you look at Figure S2D-F, you'll see a rough relationship between the coefficient values and significance as determined by permutation of the perturbation/cell assignments. 

Since the input data is transformed: log2(UMI+1),

2^(coefficient value)-1 can be interpreted roughly as the average change in UMI counts across cells due to the perturbation.

For example a coefficient value of +0.5 would correspond to an increase of 0.4 UMIs/cell for that transcript on average. 

You can try +/- 0.1 as a cutoff or 0.2 if you want to be more strict.

If it's easier, I can send you a table of signed log10(qvalues) according to the permutation test?

Does that help?

Atray

[Eduardo Gade Gusmão]

Hi Atray,

thanks so much for the explanation. I don't know why I didn't think of that before. I can do that easily. Thanks for the suggestion of the thresholds!

The only question I have left is: Why do you have the controls like "m_MouseNTC_100_A_67005" in the reg coeff table as well?

If you already have the log10(q-value) table I'll be very happy if you could send. Otherwise, there is no need to bother since I can extract the information from the regulatory coefficients.

Thank you,

Eduardo

[Atray Dixit]

I this attached file should be the FDR corrected signed log10(pvalues)

The controls were treated exactly as any other perturbation so we could evaluate the extent to which real perturbation looked different than the controls with respect to the coefficients. In the future if there are more controls in the library, as we had for Figure S5, you can reference the perturbation coefficients to the controls by subtracting the average of the controls from the rest of the perturbations. 

Atray

[Eduardo Gade Gusmão]

Hi Atray,

thank you very much, this was really helpful!

Best,

Eduardo