Perform CoV Cross-validation (hold-out)

[rcedgar]

From Artem by email: "[T]he difference in number of reads mapping between CoV+ control libraries and mammal transcriptomes is very large."

I suspect this may be misleading. If I understand correctly, Cov+ datasets have known infections by known coronaviruses, mostly (all?) Cov-19, but we are looking for incidental infections by novel coronaviruses which by definition are not in the pan-genome. Possibly, the number of viral transcripts will tend to be much lower with an incidental infection. Certainly, a novel virus will have lower identity and uneven coverage to the pan-genome compared to a positive control with a Cov-19 infection.

To model what we might see in production, Cov+ datasets should be mapped to a pan-genome with the genome of the known infection (Cov-19) excluded and of its genes plus close relatives. This is hold-out validation, also called cross-validation.

A hold-out pan-genome can be constructed using usearch as follows:

usearch -usearch_global pan_genome.fa \
-db cov19_genome.fa \
-strand both \
-id 0.95 \
-uc hits.uc \
-notmached holdout_pan_genome.fa

Here, 0.95 is the identity threshold; here all sequences having >=95% with cov19_genome.fa are removed from the reference; hits.uc is a tsv file with one record for each pan-genome sequence indicating whether it matched or not.

[ababaian]

This is very well outlined. @victorlin #27 this is a perfect example of the application of the script you're working on.