New features

[joewandy]

From slack:

so, my run gave me (maybe) some clarity...let's see what you think:
A useful system would do the following:
a) provide a single portal to annotations from lots of tools
b) standardise the scores so that it could provide a single ranking
c) Mine the annotations for other patterns (e.g. in substructures)
d) define a standard interface that would make it easy for other people to make their APIs compatible
e) be easy to install / use

As it stands, FrAnK does:
a) yes. But we probably would need to shop around and make sure we're not missing anything (new SIRIUS e.g)
b) nope.
c) nope.
d) yes, kind of
e) yessss, maybe

If I were starting from scratch now, I think b) represents an interesting and publishable scientific problem. c) would be an extension to that. We are ina good position, as we already have a framework to put a 'b' in. On the -ve side, b) feels like quite a big project so whether or not we could proceed with that would depend on people having the time / inclination.

I think this is pretty much what Rónán was saying. I just had to run around the block to work it out 🙂
Anyway, I'd be up for playing a role in b), but if the day to date practical stuff is left to me, it won't be finished before 2030.
and yes, this channel is better than the PiMP channel.

[joewandy]

ronandaly [2:35 PM]
How about combining the probs somehow?

simon [2:35 PM]
they're not really probs

ronandaly [2:35 PM]
I know
😄

simon [2:36 PM]
we thought abou this before -- some way of mapping the numbers from the different tools into the same space
that would be cool
would maybe require some solid training data

ronandaly [2:36 PM]
If we had similar chemicals to calibrate, it could be a simple regression...

simon [2:37 PM]
yes, we'd need quite a lot though
and they'd have to not come from these tools
which might be a limiting factor?

simon [2:38 PM]
anyway, back later -- will keep thinking about it.... @jjjvanderhooft maybe we should be submitting a tool that combines lots of tools to the CASMI competition

simon [2:38 PM]
that would focus the mind
(that wasn't jyst for Justin)

joewandy [2:38 PM]
a tool that combines lots of tools is the initial intention of Frank, right?

simon [2:38 PM]
yep, but to make a single prediction
and see how we got on in CASMI

ronandaly [2:39 PM]
Was thinking, if we had a score from each tool, they should be somewhat similar, map them all onto a consensus space, and hope for rough monotonicity

simon [2:39 PM]
yep, worth thinking about
and then bang that into CASMI
also, the substructure thing Justin mentioned could help there

[joewandy]

joewandy [1:23 PM]
@simon what do you mean by
Perhaps there is novelty in the MS1/MS2 combination? As a standalone tool, FrAnK would no longer have access to MS1 information from PiMP but it would be easy enough to allow it that information (i.e. a .csv file output from PiMP or some other tool (e.g. mzmine2)).
what's so novel about MS1/MS2 combination, i mean? (edited)

simon [1:25 PM]
I don't know!
but FrAnK needs some novelty
(at least according to one of the bioinformatics reviewers)
and I don't know of any tools that bring MS1 and MS2 information into the same analyses

jjjvanderhooft [1:25 PM]
the name is novel
😛

simon [1:25 PM]
true

jjjvanderhooft [1:26 PM]
there are not many that combine the two

joewandy [1:26 PM]
bring MS1 and MS2 information into the same analyses?
in what manner?
for .e.g

simon [2:13 PM]
Do we see similar expression change in things that share fragments?
My take from the reviews was that it needs some additional novelty.

What do you anticipate a user of a standalone FrAnK system using it for?

jjjvanderhooft [2:15 PM]
combining the outputs of different annotators?

simon [2:15 PM]
could be useful
combining = listing them all, or somehow merging them?
former is easy, latter is hard.

joewandy [2:15 PM]
at the moment we're just listing them all
yeah

simon [2:16 PM]
so if that's the main use case, it would surely be useful to be able to look at just the MS2 spectra that correspond to interesting MS1 features
?

jjjvanderhooft [2:17 PM]
perhaps reporting structural similarity across annotations

simon [2:17 PM]
and it would be super-useful if we worked out some mapping between the DBs used by the different tools

jjjvanderhooft [2:17 PM]
or do something that makes use of the complete list

joewandy [2:18 PM]
some mapping between the DBs?
mapping of what?

simon [2:18 PM]
Structural similarity would be cool, assuming we had structural information for all the annotations. For e.g. SIRIUS we wouldn't have
mapping = so that we know when >1 tool annotates with the same structure
at the moment that is just the user looking at the lists

simon [2:19 PM]
@jjjvanderhooft have ensemble methods ever been used in the ID competitions?

simon [2:21 PM]
Basically, I'm not sure what FrAnK is for, and that seems to be a fundamental question
It has obvious uses within PiMP
but does it do something more general outside of PiMP
?

joewandy [2:21 PM]
structure similarity above refer to what?
metabolite substructures?

simon [2:22 PM]
If two tools give different annotations but the annotations have similar structures.

joewandy [2:22 PM]
an actual structure, like SMILES?

simon [2:23 PM]
At least, what I think Justin means: if tool A says it's structure 1 and tool B says structure 2 and the two structures (SMILES if you like) share a common substructure then probably what's actually being annotated is the presence of the substructure.
all sorts of cool things there

jjjvanderhooft [2:24 PM]
exactly, or when the Tanimoto scores between SMILES is high that might give some increased confidence in the annotations

simon [2:24 PM]
but would need structures for the annotations

jjjvanderhooft [2:24 PM]
or chemical classifications

simon [2:24 PM]
which we don't currently store (maybe we do store SMILES)

simon [2:24 PM]
?

jjjvanderhooft [2:24 PM]
that can be compared across

simon [2:25 PM]
and processing to compute shared substructures
but that's a lot of news things into FrAnK