problem with generated p-values

[bitona]

Hello,

I am using Jacks to analyse a CRISPR screen where i want to select positively enriched genes compared to a control. Everything works very well in terms of beta scores, with a positive control getting the largest beta score as expected. But the p-values are the opposite of what i would expect, with our positive controls never getting low p-values. I have used as control genes both the NEGv1.txt nonessential gene list you provide and also tried to use three control genes from the TKOv3 library we use. I encountered the same problem in both. I was wondering if there was something i was missing or not using the right control genes ?

Best, Anne

[lp2]

To double check - this is a positive selection screen (hits outgrow the rest), rather than drop-out screen (hits decrease in frequency over time)? Felicity will know better, but I suspect we assume a drop-out screen in data processing, so calculate the p-values accordingly (frequency of observed or stronger drop-out). For enrichment, you should be able to use 1-pvalue instead.

Leo

On Wed, Dec 4, 2019 at 4:36 PM Anne B notifications@github.com wrote:

Hello,

I am using Jacks to analyse a CRISPR screen where i want to select positively enriched genes compared to a control. Everything works very well in terms of beta scores, with a positive control getting the largest beta score as expected. But the p-values are the opposite of what i would expect, with our positive controls never getting low p-values. I have used as control genes both the NEGv1.txt nonessential gene list you provide and also tried to use three control genes from the TKOv3 library we use. I encountered the same problem in both. I was wondering if there was something i was missing or not using the right control genes ?

Best, Anne

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/felicityallen/JACKS/issues/8?email_source=notifications&email_token=AADIQI2R5UHKZBEEKJMKUPLQW7MAXA5CNFSM4JVLN2Z2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4H6B4RQA, or unsubscribe https://github.com/notifications/unsubscribe-auth/AADIQI4O62LEQ7EIQCG4SOTQW7MAXANCNFSM4JVLN2ZQ .

[felicityallen]

What Leo said seems right to me. For the control set, more guides is better to get a good sense of the distribution of effect, so I'd suggest using more than 3 genes (I'll leave it to you to decide what are sensible controls for your experiment - anything where you're fairly confident there shouldn't be an effect should do).

Felcity

On Thu, Dec 5, 2019 at 10:47 AM lp2 notifications@github.com wrote:

To double check - this is a positive selection screen (hits outgrow the rest), rather than drop-out screen (hits decrease in frequency over time)? Felicity will know better, but I suspect we assume a drop-out screen in data processing, so calculate the p-values accordingly (frequency of observed or stronger drop-out). For enrichment, you should be able to use 1-pvalue instead.

Leo

On Wed, Dec 4, 2019 at 4:36 PM Anne B notifications@github.com wrote:

Hello,

I am using Jacks to analyse a CRISPR screen where i want to select positively enriched genes compared to a control. Everything works very well in terms of beta scores, with a positive control getting the largest beta score as expected. But the p-values are the opposite of what i would expect, with our positive controls never getting low p-values. I have used as control genes both the NEGv1.txt nonessential gene list you provide and also tried to use three control genes from the TKOv3 library we use. I encountered the same problem in both. I was wondering if there was something i was missing or not using the right control genes ?

Best, Anne

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub < https://github.com/felicityallen/JACKS/issues/8?email_source=notifications&email_token=AADIQI2R5UHKZBEEKJMKUPLQW7MAXA5CNFSM4JVLN2Z2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4H6B4RQA , or unsubscribe < https://github.com/notifications/unsubscribe-auth/AADIQI4O62LEQ7EIQCG4SOTQW7MAXANCNFSM4JVLN2ZQ

.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/felicityallen/JACKS/issues/8?email_source=notifications&email_token=AI23WQ5DYZ2US5Y62O3UONLQXDL5VA5CNFSM4JVLN2Z2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEGAJUAQ#issuecomment-562076162, or unsubscribe https://github.com/notifications/unsubscribe-auth/AI23WQ6LQV2UWWD7P3KJKELQXDL5VANCNFSM4JVLN2ZQ .

[bitona]

Thank you for your answers! Yes this is a positive selection screen. Following your advice, when i use genes with p-values equal to 1 i do obtain meaningful hits so it seems to work well (but lowering the threshold to 0.95 or even 0.99 seem to select a much higher number of genes that i suspect are false positives so i will stick to 1 for now).

Best, Anne

[lp2]

Great! I suppose p=1 means the signal is stronger than any of the controls; perhaps this is not unexpected in positive selection screens. Happy following up :)

Leo

On Fri, Dec 6, 2019 at 10:37 AM Anne B notifications@github.com wrote:

Thank you for your answers! Yes this is a positive selection screen. Following your advice, when i use genes with p-values equal to 1 i do obtain meaningful hits so it seems to work well (but lowering the threshold to 0.95 or even 0.99 seem to select a much higher number of genes that i suspect are false positives so i will stick to 1 for now).

Best, Anne

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/felicityallen/JACKS/issues/8?email_source=notifications&email_token=AADIQI3IKAVH5U2QM7MORXLQXITNDA5CNFSM4JVLN2Z2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEGDWR4Q#issuecomment-562522354, or unsubscribe https://github.com/notifications/unsubscribe-auth/AADIQI7FAAZOE3KNBYD74ILQXITNDANCNFSM4JVLN2ZQ .