[batch_int] fastmnn_feature fail

[KaiWaldrant]

Describe the bug An error occured for fastmnn_feature process on the cengen_logcp10k dataset in batch_integration full benchmark run:

Loading dependencies
Read input
Run mnn
Reformat output
Error in rbind(...) : negative extents to matrix
Calls: as ... <Anonymous> -> standardGeneric -> eval -> eval -> eval -> rbind
Execution halted

[KaiWaldrant]

After elaborate investigation I have narrowed it down but not yet confirmed.

I could not replicate the issue when running the component locally without docker. The difference between local system and docker system is the R version and the package versions:

Docker:

• R: 4.3.1
• anndata: 0.7.5.6
• Matrix: 1.6.0
• batchelor: 1.17.2
• SinglCellExperiment: 1.22.0

local:

• R: 4.2.3
• anndata: 0.7.5.6
• Matrix: 1.6.0
• batchelor: 1.14.1
• SingleCellExperiment: 1.20.1

Increasing the batchelor version on local machine did not invoke the expected error. Which would seem like the SingelCellExperiment package is the likely culprit. This has not yet been confirmed due to not being able to update R on my local system.

Further investigation/confirmation is required.