Further details about the construction of new phanta databases

[efratmuller]

Dear @yipinto and team!

I was wondering if you can perhaps share some more details about how you constructed the "UHGGv2 + UHGV "MQ+"" database you have kindly provided?

Specifically, I had 3 questions:

(1) The viral portion based on UHGV seems to also include viruses that did not meet the "MQ" (medium-quality) criteria as defined on the UHGV github. For example, vOTU-085841 is included in the phanta database but had an "uncertain" "viral-confidence" (as reported in UHGV's metadata) and therefore does not meet their "MQ" criteria. As another example, vOTU-018648 is only 49% complete (therefore not "MQ") but I see it in the database.

(2) In the UHGG2 portion it seems as though not all genomes were included (>280K genomes are listed in MGnify-UHGGv2). Can you please explain which genomes were included and how were they defined as strains/species in the db?

(3) Where were the non-bacterial/archeal/viral genomes sourced from?

Many many thanks in advance!
Efrat

[meenachakra]

Hi Efrat,
Thanks for your questions!

For point 1:

• We conducted this work based on UHGV genomes/metadata from April 2023. in our database, the only genome in the vOTU-085841 cluster is UHGV-0186730 (please see the database taxonomy/nodes.dmp and taxonomy/names.dmp). This genome is annotated as Confident viral in the UHGV metadata from April 2023.
• Similarly, the five genomes included under vOTU-018648 (UHGV-2218338, UHGV-0210983, UHGV-1933255, UHGV-1758727, UHGV-1466773) have > 50% completeness in the metadata from April 2023 and are all annotated as Medium-quality, Confident viral.

For point 2:

• We essentially dereplicated UHGGv2 at 97.5% ANI using drep and included the representatives of each 97.5% ANI cluster in our database. More specifically, we dereplicated each precalculated 95% cluster of UHGGv2 at 97.5% ANI (to reduce computational demands) and then included the representatives of the 97.5% ANI clusters in our database. The reason for choosing 97.5% was to match the ANI cutoff used by HumGut (which formed the prokaryotic portion of our original database).
• We annotated the representatives using GTDB-Tk. Each genome was assigned as a strain of its GTDB-Tk-assigned species. In the case of missing species, we assigned the species as the name of the original 95% dereplicated cluster. Please feel free to ask additional clarifying questions!

For point 3:

• UHGGv2 includes both bacterial and archaeal genomes. Viruses were from UHGV. For gut eukaryotes (fungi), we used RefSeq, and for contaminants, we used the human genome (hg38) and the Core UniVec database from NCBI.

[efratmuller]

@meenachakra Thank you for the quick reply and clear answers!

[efratmuller]

Hi @meenachakra, I had a small follow-up question: Do you think that including only species representatives (for either kingdom) rather than multiple "strains" per species as you discuss above, will dramatically decrease phanta's performance (=accuracy in resulting profiles)? Did you happen to explore more "compact" databases and compare to the detailed ones? Thank you in advance!!

[meenachakra]

Hi Efrat, we did do this for the viral portion of our original/default database and found that including all MGV strains (rather than species representatives) increased the sensitivity of Phanta to viruses. For prokaryotes, we stuck with HumGut's 97.5% threshold since they already did the work of comparing to a more compact database, please see here - https://microbiomejournal.biomedcentral.com/articles/10.1186/s40168-021-01114-w

Please let us know if you need further clarification!

[efratmuller]

Many thanks @meenachakra , I appreciate your quick responses!
Best,
Efrat