Covariate regression and ‘UNFINISHED’ error code

[Kim Hyunjung]

Hi, I’m using PLINK 2.0 for the association study. 

The disease phenotype I want to study is a binary variable. And I have 27 covariates (clinical feature), some of which are continuous, and the others are binary. When I did the association study (w/ ‘--glm’ option), I didn’t add the ‘--hide-covar’ option, so there are SNP associations with every 27 covariate in the result file. 

I have three questions about the analysis.

1. Is an association between a covariate and an SNP calculated by taking the remaining 26 covariates (plus phenotype) into account?

2. Because the phenotype is binary, PLINK did the logistic regression. However, is the SNP association with every continuous covariate automatically calculated by the linear regression?

3. Even though I got the result file, the final column ‘ERRCODE’ says that all the results are ‘UNFINISHED.’ As the PLINK 2.0 manual page says, the error code pops up when the result didn’t satisfy the usual convergence criteria when the iteration limit was hit. 

       1) I have 1184 controls and 672 diseased (phenotype) samples. Can the error message appear when the sample size is too small or because of the imbalance of the number of control samples and disease samples, like in my case?

      2) Inferred from a talk (https://groups.google.com/g/plink2-users/c/MytbMeEvlqU), I need to increase the iteration limit. Is there an appropriate method or R code to do this? 

Thanks in advance!

[Christopher Chang]

1. Yes.

2. Yes.

3a. No, that's not particularly imbalanced.

3b. No, you should instead look to remove low-value covariates.  To increase the iteration limit, you'd need to either e.g. export the data to R and perform the regression there, or edit and recompile the program; this isn't exposed as a command-line option because it's very unlikely to yield scientifically useful results.

[Kim Hyunjung]

I really appreciate the clarification. Have a nice day. 

2021년 8월 19일 목요일 오전 12시 59분 10초 UTC+9에 chrch...@gmail.com님이 작성: