Singular matrices


Congrats on developing this tool.

Sorry if this question seems naive, I’m a relatively new user.
I am trying to run the tuning function for DIABLO with two 'omics datasets which have been appropriately normalized.
During the tuning I keep getting the following errors.

Lapack routine dgesv: system is exactly singular (or) system is computationally singular

Do you have any general advice on what could be causing this and how to avoid it, if possible?

Thank you.

Thanks for using mixOmics!

There might be two reasons:
1 - perhaps you requested too many components, and after a few component iterations the residual matrices are empty.

2 - there are too many zeroes in your data (or too many missing values). Or too many zeroes in one of the Y category. Can you tell us more about the dative of data you are trying to integrate?


Thank you for the quick reply!

It is a two-block dataset of 251 mirRs and approx 3000 mRNA. The response variable is a factor class object coded as 1 or 0. I ran nearZeroVar() to pre-filter both datasets.

Either way, I resolved the issue. For some reason the keepX values were set to only a single value (5 features per component, per block) rather than a sequence, changing this resolved the issue.

Thank you.

1 Like