Difficulties running on AWS HealthOmics

[SBrennan-MBS]

Hi guys,

Thanks very much for making this tool available for all of us! I am trying to run bambu as a private workflow on AWS HealthOmics, but I am hitting a consistent error:

Warning: Bioconductor online version validation disabled; see ?BIOCONDUCTOR_ONLINE_VERSION_DIAGNOSIS [2026-06-18 01:59:02] ERROR in bambu(): error in evaluating the argument 'table' in selecting a method for function '%in%': Bioconductor version map cannot be validated; is it misconfigured?

I have disabled BIOCONDUCTOR_ONLINE_VERSION_DIAGNOSIS and even tried patching in the bioconductor config.yaml in both my R script and my nextflow.config file. From what I can tell, nextflow is passing the reference files and bam files as expected, but because I have >10 samples, a number of child R processes are spinning up and they drop both the online diagnosis and config.yaml. Then the child processes attempt online validation, can't connect to the internet, and fail.

Is there any way around this?

Cheers

[lingminhao]

Hi @SBrennan-MBS,

Thanks for considering bambu as part of your workflow on AWS HealthOmics. To confirm a few things, are you currently providing the genome as a file path, and is the HealthOmics environment effectively offline (no outbound internet access)? If so, that might explain the failure. bambu runs a quick check, BSgenome::available.genomes() when genome is provided as a path that needs online access.

One way around this is to pass the genome as an FaFile object instead of a path. When the genome isn't a plain string, bambu skips the available.genomes() lookup entirely, so no network call is needed (just make sure the .fai index also sits next to the FASTA):

library(Rsamtools)
indexFa("/path/to/genome.fa")          
genome <- FaFile("/path/to/genome.fa") # a temporary fix here
bambu(reads = bams, annotations = annotations, genome = genome)

If you can confirm this gets your run through on HealthOmics, we'll work on a proper fix for a future release so that providing a FASTA path skips the online lookup automatically.

Hope it helps!
Min Hao

[SBrennan-MBS]

Hi @lingminhao,

Thanks very much for your quick reply on this. Yeah HealthOmics only has access to AWS S3 by default so no general internet access. I have been looking into other options myself and I could set up a VPC connection so that my HealthOmics instances can have internet access, but would be very pleased if there is an easier option.

I am currently passing the reference genome in the form of a .fa.bgz, alongside fa.bgz.fai and fa.bgz.gzi. The S3 URIs for these are read from my params.json file at runtime, then nextflow reads these as channels in the workflow block:

// 3. Reference file channels
gtf_file   = file(params.gtf)
fasta_file = file(params.fasta)
fasta_fai  = file(params.fasta_fai)
fasta_gzi  = file(params.fasta_gzi)

These appear in the process block using path:

input:
path bams
path bais
path gtf
path fasta
path fasta_fai
path fasta_gzi

This path is then passed to the R script.

Should I try the following?

1. Re-build docker image, explicitly installing Rsamtools.
2. Decompress my FASTA reference genome.
3. Edit my nextflow and R scripts, passing only the decompressed FASTA file.
4. Create an index of the FASTA file during the setup steps in my R script.
5. Pass the FASTA and FAI files directly to the bambu function in the R script.

Thanks again for your help!

[lingminhao]

Hi @SBrennan-MBS,

I think FaFile reads bgzipped FASTA directly as long as the .fai and .gzi indexes sit next to it, which is exactly what you already have. Only thing you need is to open the fasta (.fa.bgz path) using Rsamtools::FaFile(), and you do not need to rebuild the docker image as Rsamtools is already part of bambu dependency, Rsamtools should be already in your docker image. Just make sure that all of the three files land in the same task directory, so FaFile can find the indexes by name.

genome <- Rsamtools::FaFile(fasta)   # 'fasta' = the .fa.bgz path nextflow stages
bambu(reads = bams, annotations = annotations, genome = genome)

Let me know if that works.

Thanks!
Min Hao

[SBrennan-MBS]

Hi Min Hao,

Using FaFile instead of directly passing the FASTA path has resolved the issue. I've now tested it on a small data subset and on the rest of my samples and it worked both times. I was also able to remove the environment variables from my nextflow config that were aimed at disabling online version checking. Thanks for your help with this!