https://nf-co.re logo
Join Slack
Powered by
We did considerable work at ampliseq and are aimin...
# ampliseq
d
We did considerable work at ampliseq and are aiming to relese 3.0.0 in the next weeks. I made a new metro map, with substantial simplifications, but otherwise it seems too crowded. Comments would be great!
Potential issues that I saw but accepted as simplification: • ONT & Illumina both default • crossover ONT to Illumina (theoretically, someone cound set for Illumina data Savont, or for ONT data DADA2, but would need explicit overwriting) • DADA2 steps heavily simplified • every single step is optional. One of DADA2 and Savont must run when raw reads are input. With ASV intermediate input, even those are omitted. However, ASV input isnt shown because most downstream stuff isnt compatible then (no abundance profile!)
@erikrikarddaniel you are right, now that you said it, the multiple boxes used for Box 3 & 4 (& 6) seem like only one of those options can be chosen, instead of many in parallel. Single dots will look not so good imho, would need connectors I think. When doing all connectors between 3 & 4, there would be a lot of noise though. Inventing a new symbol might be also not good. Ideas welcome.
I am rather moving this to the #C070369GP7T channel
v
• purple and grey color missing in legend • ITSx/ITS have two choice symbols, either missing label, or redundant? • Purple line starts and end is too much inside the box, and not connected, as opposite to green line (ANCOM) that is connected right to the choices symbols
d
indeed, good point, will try to improve that
r
Why do PacBio samples pass through DADA2? Wouldn't SAVONT be more optimal especially with the newer PacBio chemistries?
Or are the Revio Q scores the problem here?
d
Good point as well. Currently (v2.18.0) the pipeline analyses PacBio HiFi sequences with DADA2. In 3.0.0 Savont will be available in addition to DADA2. Current default in dev (=3.0.0) is kept the same, i.e. PacBio -> DADA2. However, by changing parameters, PacBio data can be analysed with Savont in 3.0.0. Since DADA2 is established and was used in the past for PacBio data, I am inclined to keep that default a little longer. (I myself didnt benchmark PacBio data analysis with DADA2 vs Savont)
r
DADA2 can take >24 hrs per my experience with revio data, wven after switching to the discrete error model
Revio Q values are binned so users that use DADA2 on default would anticipate degraded performance and spurious results
The error model learning in DADA2 is a big time sink
d
Dear Ryan, thanks for sharing your observations! Do I understand correctly that you are arguing for Savont instead of DADA2 as default for PacBio data? Could you elaborate on 
DADA2 on default would anticipate degraded performance and spurious results
? For example, it would be interesting to analyse the data once with DADA2 and once with Savont, comparing results using nf-co.re/ampliseq/dev/parameters#…
r
I think I meant the default error model learning for dada2. It struggles over both the length of the reads and, if given revio data with no change to the error modeling, dada2 will spit out an empirically incorrect error model due to the Q values from revio instruments being effectively histograms
d
default error model learning for dada2
The default doesnt apply to binned data.
if given revio data with no change to the error modeling
Did the warning added to 2.18.0 in github.com/nf-core/ampliseq/pull/1011 trigger? Binned data must be used with the appropriate settings. Or did I understand the issue wrong? What do you actually propose?
👍 1
r
@Daniel Straub oh that PR fixed the error 😅
Open in Slack
PreviousNext
Powered by