PRSice for training and test

[JuRaGa]

Hi, Thanks for developing PRSice, which is a very useful tool. I was wondering if you could me with a particular issue that I am not sure I can do with PRSice?

In order to make my results more robust, I would like to split my target population into training and test. Therefore, I would estimate the optimal set of SNPs, with the optimal P-value threshold, from the training set. Then, I would derive the PRS from the test set using the estimations from the training model. I don't know if this is possible in PRSice?

Is there an option to pass the optimal P-value threshold to PRSice, so it just returns the PRS?

Many thanks for your help, Julia

[choishingwan]

Use - -fastscore together with - -no-full and - -bar-levels

(Remove space between the -) On Fri, 22 Mar 2019 at 10:56 PM, Julia-Ramirez notifications@github.com<mailto:notifications@github.com> wrote:

Hi, Thanks for developing PRSice, which is a very useful tool. I was wondering if you could me with a particular issue that I am not sure I can do with PRSice?

In order to make my results more robust, I would like to split my target population into training and test. Therefore, I would estimate the optimal set of SNPs, with the optimal P-value threshold, from the training set. Then, I would derive the PRS from the test set using the estimations from the training model. I don't know if this is possible in PRSice?

Is there an option to pass the optimal P-value threshold to PRSice, so it just returns the PRS?

Many thanks for your help, Julia

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHubhttps://github.com/choishingwan/PRSice/issues/101, or mute the threadhttps://github.com/notifications/unsubscribe-auth/ABM44m6H5B-r9cjUFTJyt1cCy9ISO-Oiks5vZO8qgaJpZM4cDwbC.

[JuRaGa]

Hi, Thank you for your quick reply. Please could you be a bit more specific? Should I use them in the training or in the test calculation? Do those flags require any argument? Thanks again,

[choishingwan]

For the training data set, you'll just use the default parameters. For the test calculation, you can then use

--fastscore which instruct PRSice to only calculate thresholds included in bar-levels --no-full which tell PRSice to ignore p-value of 1 unless it is included in the bar-levels and --bar-levels <best threshold> with =best threshold in the training

[JuRaGa]

Hi, Thank you for your reply. I have run the test calculation with the commands you suggested, but it still included p-value of 1. Then, unfortunately the optimal PRS was the one including all SNVs, corresponding to P-value of 1, so I must have done something wrong.

I am attaching the script and the barplot and summary results, in case you can spot it? Thank you,

PRSice_Test.summary.txt PRSice_test.txt

[choishingwan]

Which version are you using? If it is 2.1.9 or 2.1.10, then I will have to have a look sometime later this week On Tue, 26 Mar 2019 at 1:20 AM, Julia-Ramirez notifications@github.com wrote:

Hi, Thank you for your reply. I have run the test calculation with the commands you suggested, but it still included p-value of 1. Then, unfortunately the optimal PRS was the one including all SNVs, corresponding to P-value of 1, so I must have done something wrong.

I am attaching the script and the barplot and summary results, in case you can spot it? Thank you, [image: PRSice_Test_BARPLOT] https://user-images.githubusercontent.com/36241035/54940238-3ea51780-4f22-11e9-98dc-15b2a95f9558.png

PRSice_Test.summary.txt https://github.com/choishingwan/PRSice/files/3004638/PRSice_Test.summary.txt PRSice_test.txt https://github.com/choishingwan/PRSice/files/3004645/PRSice_test.txt

— You are receiving this because you commented.

Reply to this email directly, view it on GitHub https://github.com/choishingwan/PRSice/issues/101#issuecomment-476297140, or mute the thread https://github.com/notifications/unsubscribe-auth/ABM44m9KCVTrTebBe-INOWOT6Rlbndr0ks5vaQVegaJpZM4cDwbC .

[JuRaGa]

Hi, Yes, I am using v2.1.9. Thanks for your help!

[choishingwan]

Could you please try again with v2.1.11 with a clean directory? As I cannot replicate this problem

[JuRaGa]

Yes, but I cannot find the link to download it. Please could you share it with me? Thanks,

[choishingwan]

You can find it here

https://github.com/choishingwan/PRSice/releases/tag/2.1.11