Singular matrix error

[nag...@oregonstate.edu]

Hello,

I'm having an issue in which SKAT produces the following error, but only for a certain combination of our phenotype and covariate files. All phenotype and covariate files combinations run successfully except for this one.

```Error in solve.default(t(X1) %*% X1) :
  system is computationally singular: reciprocal condition number = 5.3277e-17
Calls: SKAT_workflow ... resolve.list -> signalConditionsASAP -> signalConditions
In addition: Warning message:
104  samples have either missing phenotype or missing covariates. They are excluded from the analysis!

Execution halted```

The error may be the same as this one (https://groups.google.com/g/skat_slee/c/xT5tbX8hOkk/m/PVLOe4xu5aAJ) but the solution is unclear. By the sound of this thread, it sounds like there may be a fork of SKAT (a 2015 version) that avoids the issue. If so, could you please point me towards this?

Any other suggested solution or tips for troubleshooting?

Thank you to the developers and community!

[zczhao]

Hi,

It seems that there was a multi-collinearity issue in the covariates you provided and this caused that the program cannot do matrix inverse. One easy solution is to remove the covariates that cause the multi-collinearity. For example, you can delete age if you include both age and birth year as covariates.

Best,
Zhangchen