Long Read Abundances and Unbinned

[ttubb]

I have tried using CoverM to calculate relative abundance of MAGs in a metagenome sample. After reading the documentation and inspecting results, I am wondering how suitable this is for my long read data. From the README.md

Formula: 0.02235294/0.02235294*(2/6)
Explanation: If the contig is considered a genome, then its mean coverage is 0.02235294. There is a total of 0.02235294 mean coverage across all genomes, and 2 out of 6 reads (1 out of 3 pairs) map. This coverage calculation is only available in 'genome' mode.

The MAG abundances are scaled based on the fraction of reads mapping to any MAG. Intuitively, I expected number of read bases instead of read count for this scaling (in case of a long-read dataset). I am scared of read-length-biases between assembled and unassembled fractions. Do you have any experience on whether this makes a meaningful difference? Are you open to including a flag that changes the scaling approach?

And while I am suggesting features: When analyzing datasets in our group, the total abundance of the non-binned contigs is a metric we always check and report. This can of course not be calculated from coverage, but needs to happen based on read or base count. For my current dataset I did this with a custom script after running CoverM. But of course it would be nice to provide a .fasta with unbinned contigs and get this data directly from CoverM.

Thank you for any help and of course for providing this tool,
~Tom

[AroneyS]

That is a good point. I can look into either altering this method or adding another one using read-lengths. In the meantime, choosing mean (or trimmed_mean) and calculating relative abundance yourself should work fine.

By total abundance of unbinned contigs, do you just mean proportion of reads (or bp) that map to them?

[ttubb]

Thank you for the quick reply! I did what you suggested (calculated the relative MAG abundances based on trimmed mean and then scaled by fraction of read bases).

An example of what I had in mind regarding unbinned contigs:

name	abundance
MAG_A	13 %
MAG_B	22 %
MAC_C	15 %
unbinned contigs	30 %
unassembled	20 %

For my current dataset I calculated the values for unbinned contigs and unassembled based on the number of bases in the respective reads. The distinction is important for downstream tasks (e.g. metaproteomics).