STARsolo custom reference fails during genomeGenerate — need genomeChrBinNbits for fragmented Nicotiana genome

usegalaxy.eu support
, , , ,
1

Hi Galaxy Europe team,

I am trying to reconstruct a published 10x Genomics v3 single-nucleus RNA-seq dataset from Nicotiana tabacum K326 using RNA STARsolo on singlecell.usegalaxy.eu.

Tool: RNA STARsolo 2.7.11b+galaxy2

I am using a custom K326 reference:

  • genome size: ~3.75 Gb
  • 382,373 FASTA sequences/contigs
  • matching GTF annotation
  • 10x 3′ v3 FASTQs

The job fails during STAR genome generation, before alignment.

The generated command includes:

STAR --runMode genomeGenerate \
  --genomeDir tempstargenomedir \
  --genomeFastaFiles refgenome.fa \
  --sjdbOverhang 100 \
  --sjdbGTFfile ... \
  --genomeSAindexNbases 10 \
  --runThreadN 4 \
  --limitGenomeGenerateRAM 40960000000

STAR exits with:

EXITING because of FATAL PARAMETER ERROR:
limitGenomeGenerateRAM=40960000000 is too small for your genome

SOLUTION: please specify --limitGenomeGenerateRAM not less than
271935602954 and make that much RAM available

I previously tested genomeSAindexNbases values 14 and 13, and now 10. The required RAM remains approximately 272 GB, so changing that parameter does not solve the problem.

The reference is extremely fragmented, and STAR recommends reducing genomeChrBinNbits for genomes with many reference sequences. Using the STAR recommendation for this assembly gives approximately:

--genomeChrBinNbits 13
--genomeSAindexNbases 14

However, the Galaxy STARsolo wrapper exposes genomeSAindexNbases but not genomeChrBinNbits.

Would it be possible to help with either of these options?

  1. Expose --genomeChrBinNbits for custom-reference genome generation in RNA STARsolo; or
  2. Build/register this K326 STAR index with genomeChrBinNbits=13 so that it can be selected as a built-in index.

Failed job ID: 4838ba20a6d86765bfc0340674b062e5

Shared history:
singlecell.usegalaxy.eu/u/fforn/h/chen-sgt-lgt-tobacco

The shared history contains the failed job, workflow inputs, reference FASTA and GTF.

Thanks in advance for your help.

2

Hi @fforn
Congrats on posting your first question on GalaxyHelp!!

I think the easiest solution is to build an index for your reference genome. I will forward this to Galaxy Europe admins and post an update here when the index is ready.

Best,

Pavan

3

Hi again,

The reference genome version you are using is from 2014 and is still at the scaffolding level. The annotation also seems incomplete.

There is a latest assembly from 2023 on NCBI with a new annotation from 2025.

Does it still fit your research question, or should it be the older genome from 2014?

4

Hi Pavan,

Thanks a lot for your comment and for forwarding this to the Galaxy Europe admins. The corresponding author sent me the exact GTF used for the Cell Ranger analysis, and following your suggestion I found a chromosome-level K326 FASTA that matches it; gffread successfully reconstructed all 63,697 transcripts.

It looks like my main mistake was starting from a very old, highly fragmented scaffold-level assembly. So I definitely learned that, when working with older plant genomes, it is worth checking the assembly level and FASTA/GTF compatibility before trying to solve STAR indexing problems with more RAM or parameter tuning.

STARsolo has now been running for over an hour instead of immediately crashing with the ~272 GB RAM error, so this looks much more promising. If it works, the custom index for the old reference probably won’t be needed.

Thanks again for the help!

Federico

5

Great! Hope it will work out. If there is any other issue, please let us know.