DEXSeq error

[xguo]

Hi, there,
I have been trying to use DEXSeq for alternative splicing analysis. We have RNA-Seq data for 9 pairs of normal/tumor samples and I'm doing paired analysis following example in the DEXSeq manual. Some errors I got when doing testGeneForDEU are as follows:

Error in chol.default(XVX + lambda * I) :
the leading minor of order 22 is not positive definite
Error in chol.default(XVX + lambda * I) :
the leading minor of order 22 is not positive definite
In addition: Warning message:
In testGeneForDEU(rumExons, formula0 = formula0, formula1 = formula1, :
Error in fit1 for gene ENSG00000127928, exon E008: Error in chol.default(XVX + lambda * I) :

Does anyone what these errors mean?

thanks a bunch!

[Wolfgang Huber]

Dear Xguo

Can you make sure that you are using a current version of DEXSeq (e.g., latest release http://www.bioconductor.org/packages...ml/DEXSeq.html)

If the error persists, can you please post

• the output of sessionInfo().
• a minimal version of your R script that reproduces the error.
• ideally, the input data that is necessary to run the script and reproduce the error.

Best wishes
Wolfgang

[xguo]

Yes, I'm using the current version from bioconductor. I'll send some input data for you to reproduce the error.

thanks for looking into it.

best

[areyes]

Hi xguo,

Thanks a lot for your feedback!

I gave a closer look to the error. The problem arises when calling glmnb.fit at the time of testing. If you look at the count values where the function fails, you will see that these are strange cases e.g. in exon E033 where the variance is really high.

[CODE]
> countTableForGene(rumExons, "ENSG00000007350")["E033",]
1N.txt 2N.txt 1T.txt 2T.txt
0 1 73 0
[\CODE]

These type of exons are not very likely to be called significant, so it should be harmless for your analysis. For the moment, DEXSeq version 1.1.3 now converts this error into a warning (just to avoid your screen to be full of this errors). Anyway, I will try to look for a smarter solution.

How often do you see this error in your complete data set?

[xguo]

Thanks a lot for looking into the error and finding out the cause. As you said, it happens when the variation is too large. It doesn't happen very often.

Do you have any plan to include reads mapped to splice junction in your analysis? DEXSeq currently uses reads mapped to exons only for differential exon usage testing. It seems to waste a lot of information by ignoring junction reads. It'll great if junction reads can also be included in the model or tested separately for cross-validation with exon reads.

thanks a lot
Xiang