TaxVamb - KeyError: "Separator 'C' not in sequence identifier

[AnthonyC-S]

I wanted to try out TaxVamb for binning one metagenomic sample. But ran into this error after training. Here is the log.txt.

I ran TaxVamb with this command (excluding slurm parameters):

vamb bin taxvamb --outdir /workdir/awc93/abx-meta/taxvamb_output/seqid_14 \ --fasta /workdir/awc93/abx-meta/spades_output/before_treatment/JAX/seqid_14/contigs.fasta \ -m 250 \ --abundance_tsv /workdir/awc93/abx-meta/strobealign_output/seqid_14_abundances_250_minlength.tsv \ --taxonomy /workdir/awc93/abx-meta/taxvamb_input/seqid_14_taxvamb_tax_input_250_minlength.tsv

the contigs were generated with metaspades, and the abundance_tsv was made with strobealign with this command:

strobealign -t 30 --aemb /workdir/awc93/abx-meta/spades_output/before_treatment/JAX/seqid_14/contigs.fasta /workdir/awc93/abx-meta/reads_postqc/before_treatment/JAX/seqid_14/seqid_14.trim_1.fastq /workdir/awc93/abx-meta/reads_postqc/before_treatment/JAX/seqid_14/seqid_14.trim_2.fastq > seqid_14_abundances.tsv

and the taxonomy was done with Kraken2 and then converted with taxconverter.

the nodes in the contigs.fasta file are named in this manner:

NODE_1_length_224708_cov_8.543349,
NODE_2_length_175860_cov_10.485417
etc.

Python version: 3.10.16
Vamb version: 4.1.4.dev144+g95f1155

How should I be running TaxVamb to avoid this error? Should I be running multiple samples instead of doing a test run with a single sample? Thank you for any help!

[jakobnissen]

This appears to be a bug in clustering - or rather, in the bin splitting. I'll look at it later today.

[AnthonyC-S]

Thank you! I re-ran it but this time included -o "" in the vamb CLI parameters. It successfully ran and here is the log:
log_successful_run.txt. Would I get better results including bin splitting?

[jakobnissen]

I have a fix in #377. Thanks for the bug report.
Yes, binsplitting normally give better results, since there will often be variation between samples. Assembling each sample individually, then binning them together (to make use of co-abundance across samples), and then splitting to yield sample-wise pure bins usually gives the best result.