Details for people who like phylogenetics: virological.org/t/inter-clad...

I think you just discovered a meme

To mitigate mapping compute time vs output, we suggest making PVGs using population structure info.
This means 1 sample from each main lineage for a multi-sample PVG.
This can be combined with SNPs from a 1-sample PVG to avoid coordinate confusion & to allow interpretation by non-experts.
[7/7]

We looked reads (red/blue in image) mapped to our 3-sample PVG (3 genomes shown in black, dark grey, light gray).
We found they mapped better to one PVG path at certain regions.
This illustrates the biases inherent in using a single linear reference genome, which misses ~27% of SNPs.
[6/7]

For example in the 3-sample PVG, SNPs were detected right across each of the 3 genomes (image).
Giraffe found additional SNPs when mapping compared to Minimap2, even to our 1-sample PVG.
Overall, 27% of 3-sample PVG SNPs could not be projected onto the linear reference genome.
[5/7]

We mapped all available short read libraries to these LSDV 1-, 3-, 6-, 121-sample PVGs with #VG #Giraffe & compared that to mapping to a linear reference with #Minimap2. Generally, most SNPs were found by both methods (black in image), but Giraffe found more SNPs (blue) and missed few (red).
[4/7]

The 3-sample PVG captured 94% of the nodes of the 6-sample one.
Similarly, the 6-sample PVG retained 82% of the 121-sample PVG.
The 3-sample PVG had 1 sample from each of the main clades, so we tested if this was sufficient for short Illumina read mapping (relative to the 121-sample PVG).
[3/7]

We created bigger & smaller pangenome graphs (1, 3, 6 and 121 genomes) using our tool Panalyze (github.com/downingtim/P...) which is essentially a wrapper for #PGGB, #OGDI, #Bandage and related tools.
This meant we could analyse different PVGs quickly.
[2/7]