Sleuth Error: System is computationally singular

Question

[scottschu97]

Does anyone have the solution to this error message?

I am working with Kallisto and Sleuth to analyze some RNA seq data. I have a control set of macula data and a set of macula data with AMD. I am trying to analyze differential gene expression between the two sets.

 Design table:
 sample Tissue  Condition
 AMD_macula.11  Macula  AMD
 AMD_macula.12  Macula  AMD
 AMD_macula.14  Macula  AMD
 AMD_macula.17  Macula  AMD
 AMD_macula.18  Macula  AMD
 AMD_macula.19  Macula  AMD
 ctrl_macula.10 Macula  nodisease
 ctrl_macula.13 Macula  nodisease
 ctrl_macula.15 Macula  nodisease
 ctrl_macula.16 Macula  nodisease
 ctrl_macula.4  Macula  nodisease
 ctrl_macula.6  Macula  nodisease

I have created my sleuth object and now I am trying to fit the models so I can run linear regression and wald tests.

 so <- sleuth_fit(so, ~sample + Condition,'full')
 so <- sleuth_fit(so, ~Tissue, 'Tissue')
 models(so)

When running these, I cannot get past creating the first model because I am getting this error that is saying that the system is computationally singular.

 > so <- sleuth_fit(so, ~sample + Condition,'full')
 Error in solve.default(t(X) %*% X) : 
 system is computationally singular: reciprocal condition number = 3.82836e-18

Does anyone know how to solve this? I believe something is wrong with my design table, but I cannot think of how to fix it.

Thanks!

[biofalconch]

It seems that there is a problem with the code. And I see you already posted something over Sleuth's google group. I think the best option here is to wait for the sleuth's development team to give you some advice

Answer 1

[Brian Gudenas]

Is "sample" just a unique sample identifier?? Then it should not be included as a batch variable and you should drop it from your sleuth_fit call. Also why are you including tissue if they are all from the same tissue, if there is no variability then there is no point in including it as a batch variable.. A meaningful batch variable should be a factor variable with multiple samples per factor.

Try:

so <- sleuth_fit(so, ~ Condition, 'full')
so <- sleuth_fit(so, ~ 1, 'reduced')

[scottschu97]

Thank you! This helped!

If I did have samples with different tissues as well should I call:

so <- sleuth_fit(so, ~ Tissue, ~ Condition, 'full')
so <- sleuth_fit(so, ~ 1, 'reduced')

[GenoMax]

If an answer was helpful you should upvote it, if the answer resolved your question you should mark it as accepted.

[Brian Gudenas]

The full model should only be testing the effect of condition not Tissue + condition, so you would need to add Tissue to the reduced model.

so <- sleuth_fit(so, ~ Tissue, ~ Condition, 'full')
so <- sleuth_fit(so, ~ Tissue, 'reduced')

[scottschu97]

Thank you! The help is much appreciated!