Data imputation

[Adafede]

Small summary of what has been discussed with @oolonek

Imputation of missing data can go multiple ways:

1. LOTUS structures attributed to LOTUS organisms (level beginner)
2. LOTUS structures attributed to non-LOTUS organisms
3. non-LOTUS structures attributed to LOTUS organisms
4. non-LOTUS structures attributed to non-LOTUS organisms (level hardcore)

Structures/organisms include of course also their higher parents.

Possible limitation:
While the relatively well-defined biological tree of life will guide imputation on the side of the organisms, the lack of an equivalent tree on the chemical side might be an issue.
If we want to look for neighbor structures, let's say in PubChem we will land on many synthetic compounds we do not want to start imputing to the organisms. This won't be the case for a non-LOTUS studied organism as we know it is a living organism. On the chemical side, much harder.

[oolonek]

Added numbers so we can ref to these in the discussion. Also if their is a gradient from beginner to hardcore, I shifted LOTUS structures attributed to non-LOTUS organisms from initial 3. to 2. I feel it comes with less issues than 3.

[oolonek]

Another thing that comes to my mind when having a look at this is that here we look this at the most detail level :
structure <-> biological organisms

However we will most likely never reach such precision, or at least not in the first moves, and rather work at:
chemical classes <-> biological taxa (>= genus)

[oolonek]

Regarding point 3. and the comment above.
Do we know if we have ALL the NPClassifier taxa in lotus ?
Maybe starting by the imputation of missing NPClassifier taxa could be a first step ?

[Adafede]

Very interesting point, I'll assign it to myself.

[Adafede]

Done, see https://github.com/anticipated-lotus/elotus-trees/blob/main/inst/scripts/count_npclasses.R and #9 (comment)