How to harmonize output

[SilasK]

dbcan perform 3 annotations: Diamond, eCAMI, and HMMR. The output of the tree don't necessarily agree. Do you have a preferred way to harmonize them?
Should we trust HMMR more? Is DIAMOND really reliable?

I know in the publication you had different comparisons, but this was before eCAMI was added.

[SilasK]

https://twitter.com/SilasKieser/status/1574817427208245249

[yinlabniu]

In most cases, they agree. For those not, we do have an internal way to harmonize. The preferred order is HMMER, eCAMI, Diamond, meaning if HMMER produces output, we take its annotation over the other two; if HMMER does not have output, we take eCAMI annotation. In this order, we basically ignore the Diamond annotation. However, we do need Diamond for the 2/3 voting. Yes, this was done on Hotpep before eCAMI came out. But the prioritization strategy is inherited. Yanbin

…

________________________________ From: Silas Kieser ***@***.***> Sent: Wednesday, September 28, 2022 2:13 AM To: linnabrown/run_dbcan ***@***.***> Cc: Subscribed ***@***.***> Subject: [linnabrown/run_dbcan] How to harmonize output (Issue #106) Non-NU Email

________________________________ dbcan perform 3 annotations: Diamond, eCAMI, and HMMR. The output of the tree don't necessarily agree. Do you have a preferred way to harmonize them? Should we trust HMMR more? Is DIAMOND really reliable? I know in the publication you had different comparisons, but this was before eCAMI was added. — Reply to this email directly, view it on GitHub<https://urldefense.com/v3/__https://github.com/linnabrown/run_dbcan/issues/106__;!!PvXuogZ4sRB2p-tU!FAtDGfHq0NOVMK3mI1MgjshJYb9QFR3dKxUF_3FDx6X7Xnp0i-d8kYFgeReGHSnYnd5iU-2xnwHW585bEEIXuw$>, or unsubscribe<https://urldefense.com/v3/__https://github.com/notifications/unsubscribe-auth/AEXNKZTD2EKY6JNF2I5RSHLWAPV2PANCNFSM6AAAAAAQXO57T4__;!!PvXuogZ4sRB2p-tU!FAtDGfHq0NOVMK3mI1MgjshJYb9QFR3dKxUF_3FDx6X7Xnp0i-d8kYFgeReGHSnYnd5iU-2xnwHW584dONwLLw$>. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[SilasK]

Thank you for the clarification. Would you still keep diamond prediction for genes that are not annotated with the other two tools? And if yes to down to the default E-value threshold?

Also if Hmmer predicts one annotation and DIamond 2 you wouldn't keep the second from diamond.

[yinlabniu]

No, by default, we remove predictions resulted from just one tool (including those only predicted by diamond). If hmmer and diamond have different annotation, we keep hmmer's.

…

________________________________ From: Silas Kieser ***@***.***> Sent: Thursday, September 29, 2022 3:02 AM To: linnabrown/run_dbcan ***@***.***> Cc: Yanbin Yin ***@***.***>; Comment ***@***.***> Subject: Re: [linnabrown/run_dbcan] How to harmonize output (Issue #106) Non-NU Email

________________________________ Thank you for the clarification. Would you still keep diamond prediction for genes that are not annotated with the other two tools? And if yes to down to the default E-value threshold? Also if Hmmer predicts one annotation and DIamond 2 you wouldn't keep the second from diamond. — Reply to this email directly, view it on GitHub<https://urldefense.com/v3/__https://github.com/linnabrown/run_dbcan/issues/106*issuecomment-1261911620__;Iw!!PvXuogZ4sRB2p-tU!A2HUWMAZt_YCPdWn6Th7kjuajJaNLBTY09AkNnvSffxaCELih8dW4iDRNslIsf9O-3-6ZxS7lDiYGMZQzEud-A$>, or unsubscribe<https://urldefense.com/v3/__https://github.com/notifications/unsubscribe-auth/AEXNKZVLKGLAJEUNVRKCWVLWAVEHXANCNFSM6AAAAAAQXO57T4__;!!PvXuogZ4sRB2p-tU!A2HUWMAZt_YCPdWn6Th7kjuajJaNLBTY09AkNnvSffxaCELih8dW4iDRNslIsf9O-3-6ZxS7lDiYGMZDEtsqeQ$>. You are receiving this because you commented.Message ID: ***@***.***>

[SilasK]

Ok, you keep annotations that are predicted by two tools. I saw this gave the best performance shown in the publication.
This means you also drop annotations that come from only HMMER?

[yinlabniu]

that's right. if only hmmer generates output and the other two are not, we drop it. But users can decide to keep it since you have the raw results.