set fixed p value threshold

[Tomas Keller]

Hi Sam and others,

I have a question regarding the p value thresholds:

I want to set a fixed P value threshold, say p=0.05, to compute a best PRS, but if setting 

> 

--lower 0.05 \

--upper 0.05 \

The best PRS is derived by p=1 instead of p=0.05. Also the bar plot showed for p=1. 

Is p=1 a default setting? 

Thank you.

Tomas

[Sam Choi]

you will need to use --no-full

Best way to do it is

--no-full --bar-levels 0.05 --fastscore

[Tomas Keller]

Hi Sam, 

thanks very much!

If the best result (say highest R2) obtained by setting p value threshold p=0.1 (not other p value thresholds)  from the discovery data (base file), 

we should then use p=0.1 for the target file as well. But obviously people are doing differently. What's your opinion? 

Thomas

[Sam Choi]

It is sometimes confusing with these terminology:

Base is the summary statistics, and it should only be used for the effect size estimates and should be independent from the target data

Target data is where we usually perform PRS on. Generally, people will estimate the best p-value threshold from it and just report the R2

However, it will be best to use a separate independent data set call the validation data set to estimate the R2 to avoid overfitting and stuff.

As a result, I am not completely certain what you mean by getting the highest R2 in the base file (as you will still need the genotype to calculate the R2 and stuff)

[Tomas Keller]

Hi Sam, thank you very much for your explanation. 

What I meant is that: target data should be 'validation data set' and the best p value threshold is from the base (either cross validation or leave one dataset out approach.

In this case, the target sample can obtain a fixed risk score no matter what kind of phenotypes of interest are used (height, BMI, ...).