Gene and Drug Pearson correlation

Q&A
,
1

Dear DepMap team,

Would you please help me to understand gene and drug pearson correlation on selected group of cell lines?
For the circled 3 cell lines, it shows that they have similar drug sensitivity to CB-5083, but AURKA gene is most essential to the cell line at the lowest position. Over all, all the cell lines in the image are negative correlated with CB-5083, this means that the more essential AURKA gene is to a cell line, the less sensitive this cell is to CB-5083, am I correct? Thank you!
image

Best,
Marissa

2

Dear Marissa,

Your interpretation is correct based on the plot but it is tricky to interpret the negative correlation between two vulnerability phenotypes! I would be particularly hesitant to put too much weight on your case since all the numbers are very low, i.e. all 13 cell lines you pointed are essentially dead either under CB-5083 treatment or AURKA knock-out and it is hard to say who is more dead.

Warmly,
Mustafa

3

Hi Mustafa,

Thank you for the answer. The reason that you said all 13 cell lines were dead either under CB-5083 treatment or AURKA knock-out is because LFC is negative and CRISPR KO effect is also negative?

Also, is the LFC measuring the cell numbers in treatment group and DMSO group? For example, LFC is 2 means that there are 2 fold of more cells in treatment group than DMSO group, while LFC is -2 meant that there are 2 fold of less cells in treatment group than DMSO group?

Thanks,
Marissa

4

That is correct, and in your figure LFC values are all below -3, which means they are at least 8 fold less than the control. I believe that is pretty close to the dynamic range of the assay.

5

Thanks Mustafa! I am looking for sensitive CRIPSR genes and drugs for these 13 cell lines, what are the dynamic range would you recommend for CRIPSR genes KO effect and drugs LFC?

Thanks,
Marissa

6

For drug LFC’s I would say anything below -1.7 (corresponds to 30% viability) is a good rule of thumb. For KO’s, I would use dependency probabilities rather than gene effect and 0.5 is a good cut-off for them.

7

Hi Mustafa, are you saying taking drug LFC in PRISM database from 0 to -1.7, or from -1.7 to -3 as sensitive drugs?

For KO’s, I downloaded CRISPR data from CRISPR (Avana) DepMap Consortium 21q1, the number for each cell line is KO effect. Where can I find dependency probabilities for these cell lines? Should I take any genes with dependency probabilities higher than 0.5 as sensitive genes?

Thanks,
Marissa

8

Apologies for the confusion.

You can use PRISM LFC values less than -1.7 as sensitive drugs.

For the Avana dataset, you can download " Achilles_gene_dependency.csv" from the downloads section and interpret any gene >0.5 as a dependency.

9

Thanks Mustafa. You mentioned that LFC values are all below -3, which means they are at least 8 fold less than the control and that is pretty close to the dynamic range of the assay. Does it mean I should only take the drugs with LFC between -1.7 to -3. For example, there are 2 drugs A and B: A with LFC -3, B with LFC -7, can I say B is more sensitive than A to my cell lines?

Thanks,
Marissa

10

i would say drug B is more effeective than drug A but I would be hesitant to say the difference is 16 fold (2^(7-3)) since the assay is not designed to differentiate so small abundances.

11

Thanks Mustafa. There are CRISPR gene dependency file and Achilles_gene dependency file, what are the differences between them?

Thanks,
Marissa

12

Hi Marissa, the CRISPR files are the result of integrating Achilles and Project Score data using the methodology described in https://www.nature.com/articles/s41467-021-21898-7. You can find more details in the dataset README.