Unexpected BGC Count and Duplication Issues in BiG-SCAPE v2 Analysis

[libradaatencio]

Hello everyone,

I am analyzing a set of BGCs identified using AntiSMASH v7, consisting of 1,524 .gbk files. In my last run, I executed the following command:

bigscape cluster -i BGCs_input/ -o output_28ene2025 -p pfam/Pfam-A.hmm --classify class --mix --alignment-mode auto --gcf-cutoffs 0.4,0.5,0.6

I used these parameters for the following reasons (See screenshot below):

--classify class: Provides more detailed classifications of the BGCs in the bins.
--mix: Generates a general network including all BGC files.
--alignment-mode auto: Ensures proper alignment since some detected BGCs may contain a contig-edge tag.
--gcf-cutoffs 0.4,0.5,0.6: These cutoffs were chosen based on previous runs to avoid overestimating clusters.

Based on my results, I have the following questions:

1. Why is the total number of BGCs (2,495) higher than the input count (1,524)?
2. How can I prevent or minimize the duplication of BGCs across different bins? Should I use the --hybrids-off parameter? I noticed this parameter seems to work differently compared to BiG-SCAPE v1.
3. Regarding genome count, the results indicate only one genome analyzed, but I am working with 32 fungal genomes (2 genomes sequenced in our lab + 30 representative genomes from GenBank). Should I modify the file name? Is there a recommended approach to address this?

Thanks in advance for your help!

[nlouwen]

Hi @libradaatencio,

Thanks for reaching out!

1. It looks like we are counting some BGC records twice since they occur once in the mix set and again once in the classify set. Thanks for pointing that out, we'll be able to easily fix this.
2. There will be much overlap between mix and classify: first comparing all BGCs together in one mix bin and then classifying based on antiSMASH class. However, there should not be any duplicated BGCs within or between the classify bins (if that is the case, please let us know). Using the --hybrids-off parameter allows you to add e.g. an NRPS.T1PKS BGC to both the NRPS and T1PKS bins instead of a NRPS.T1PKS bin. This essentially duplicates this BGC between those two bins, but can be useful if you want to analyze all antiSMASH classes separately regardless of genomic context.
3. So far, the most reliable method we've found for counting genomes is to parse the ORGANISM header in the .gbk files. However, we've also noticed that this is not always filled. Could you check if that is the case for you input .gbks? If they are not filled, they will likely be grouped into Unknown, leading to 1 present genome.

Let me know if you any further questions!

[libradaatencio]

Hello @nlouwen
Thanks for your answer.

I performed 2 "experiments" following your suggestions, I think both runs were succesfull:

Experiment 1:
bigscape cluster -i BGCs_input/ -o output1_06feb2025 -p pfam/Pfam-A.hmm --classify class --alignment-mode auto --gcf-cutoffs 0.3,0.4,0.5,0.6

Experiment 2:
bigscape cluster -i BGCs_input/ -o output3_06feb2025 -p pfam/Pfam-A.hmm --classify class --alignment-mode auto --mibig-version 3.1 --gcf-cutoffs 0.3,0.4,0.5,0.6

In both cases, I did not use the mix parameter. As you mentioned, the ORGANISM header was empty in the input files. After fixing that, BiG-SCAPE was able to correctly recognize all 32 genomes included in the analysis. So, thanks for the suggestion!

For the second experiment, I added --mibig-version 3.1, and the run completed successfully. I now just need to carefully review my results. They included more genomes and, consequently, more BGCs.

I think that the --gcf-cutoffs 0.3,0.4,0.5,0.6 will allow me to determine the best run to present in my results. So far, it seems that the optimal cutoff falls between 0.4 and 0.5.

Thanks again for your suggestions.

Librada