Adding custom amplicon panels

[ggrimes]

I am using a custom amplicon panel for wastewater sequencing and would like to know the process of adding this to the primer sets folder.

E.g. The naming conventions needed and how to go about generating the amplicon_distribution.tsv

Thanks for any help.

[willboulton]

Hi, thanks for submitting this issue. At the moment, the way to do this isn't particularly 'sleek'. If you take a look at the wiki (https://github.com/goldman-gp-ebi/sars-cov-2-metagenomic-simulator/wiki/SWAMPy-method), section 1 under user chosen input files gives a description of the format for the bed, fastq, and amplicon_distribution files you'd need to add for your own primer set. The names of the files can be arbitrary. Together the bed file and the fastq file provide the definitions of the primers - the quality scores in the fastq file are just dummy scores so really the only important information in there are the primer names which need to be formatted correctly (labelled as left and right, e.g. @primer_1_LEFT or @primer_1_RIGHT) and the primers themselves.

The difficult part is setting the amplicon_distribution file, and also tuning any of the other parameters that might be relevant for you, based on experiments. The column 'hyperparameter' is just the relative abundance of that amplicon, so this column should sum to 1. To set this parameter for other primer sets, we looked at relative amplicon frequencies across a large number of experiments. We use the variance across experiments to set the --amplicon_pseudocounts parameter, in the wiki we suggested a deafult value of 10000 for this but this is probably too high and would produce extremely consistent relative abundances of amplicons between simulations - this might require some tuning for your situation, but a value of about 200 might be suitable. There are lots of other parameters to do with PCR error frequencies, (listed here: https://github.com/goldman-gp-ebi/sars-cov-2-metagenomic-simulator/wiki/CLI-arguments) which have default values derived from experiments, but might be different for your primer set. In the first instance it might be easiest (especially if you are not concerned with PCR errors) to use the option --no_pcr_error, and add these errors in once you're satisfied with the results from the simulator with only sequencing errors introduced.

The final step would involve some manual code editing - you would need to edit the file src/simulate_metagenome.py (https://github.com/goldman-gp-ebi/sars-cov-2-metagenomic-simulator/blob/main/src/simulate_metagenome.py) in 4 places:

• line 62 (with a primer set option). Add a code for your primer set under where you see the option 'choices'.
• lines 154, 182, and 202 - copy and paste the above few lines of code but adding your file names and using the code for (i.e. following the same pattern as for the other choices a1, a4, n2).

Hope these instructions are useful, otherwise happy to help further. If you get your primer scheme working, we would be very glad if you are willing to submit a pull request!

[EvaFriederike]

Hi, thanks for opening this issue!

I want to simulate spike-only sequencing data using the ARTIC v4 panel and currently have three options:

1. Simulating whole-genome with all default settings and filtering only those reads that map against the spike region. Here I might have to increase n_reads in order to yield more depth for my target region.

2. Adjusting the amplicon distribution by setting the proportions for amplicons outside the spike gene to zero and distributing proportions across the remaining amplicons accordingly.

3. Simulating spike-only by using only the spike-covering ARTIC v4 primers as a custom input primer panel (using your tutorial for adjusting the code worked like a charm :) ).

Which version would you advise me to go with?
And in case it is option 2. or 3., how would you advise me to handle the PCR error option and the amplicon_pseudocounts parameter in this case? Can I use the default setting and allow PCR error simulation since I basically use one of your provided default primer schemes that you used to determine experimentally derived default parameter settings? Or does the fact that I only use a subset of the primers change the impact of those settings?

Cheers
Eva

[willboulton]

Dear Eva,

Thanks for your message - I am very glad that you can make some use of our simulator!

If you have managed to get option 3. working - I would suggest that as best.

I think using the default PCR error rates is ok - we treat sites as essentially independent and uniform when estimating these rates (other than amplicon overlaps). You can take a look at our preprint https://www.biorxiv.org/content/10.1101/2022.12.10.519890v1 and perhaps decide if you think our model is good enough for your use case. Otherwise, perhaps if you want a feature to be implemented we can see if it's possible.

For amplicon pseudocounts, do you have any experimental data to work with? If not, this is somewhat subjective as you need to decide how 'noisy' you want your amplicon counts to be per simulation, compared to the amplicon distribution. Setting this to 200 produced 'high quality' (i.e. consistent) results compared to our experimental data, whereas for 'low quality' samples it could be closer to 20. But if you don't know and just want some test data for a simulation experiment, perhaps leaving this at 200 is fine.

Using a different primer set probably does impact some settings so it is really a case of how much modelling inaccuracy you mind having, and if you have an alternative option. Your option 3. + using all default parameters should produce realistic-ish simulations relatively quickly and easily. We also did a set of simulation experiments in our preprint, and if you wanted you could use the parameters we used for that (details are in the supplementary information of the preprint). If you have some real data, you can then think about tuning parameters to fit that better - perhaps there will also be other features in your data that we haven't modelled but which are relevant for you... in which case we would like to know, if you are happy to tell us! It would be great to be able to improve this simulator.

Best wishes,
Will

[rabiafidan]

Hi Eva,
Thank you for your question!

I just wanted to add about PCR errors, we first create a PCR error reserve with a genome-wide approach, irrespective of the data provided in that run. And only after we have that reserve, they are applied to the data (mutate the amplicons). So if you exclude some regions, PCR errors that were supposed to be in that region would just do nothing. It wouldn't cause things like a much higher density of PCR errors in the remaining regions if that is your concern. And yes, as Will pointed out take a look at the options we found useful in the preprint! You might want to increase the default PCR error rates a bit as we did.

And I agree with Will, the ideal option would be 3. But depending on how many simulations you will run and your target read depth (i.e. how computationally intensive it is going to be) 1st option might be acceptable and probably the easiest. The 1st option will assume when you sequence only spike region (assuming using only a subset of the primers in laboratory) will not affect the relative abundance distribution of the remaining amplicons.

All the best,
Rabia