federatedcloud / douglasmetabolicmodels Goto Github PK

• Come up with alternative analysis results with inorganic metabolites removed. We still want to keep the current tables being generated as-is, as well.
  • HeatMaps
  • overlap summary
  • CMP (individual.csv)
  • CMP (differential.csv)
• Generate random removals of metabolites (separate analyses for 1 to 5 metabolites); start with new rich model, then remove all uptake reactions that ARE NOT part of the minimal reaction bounds.
• Add community growth rate and res.BM to the heat maps.
• Add an efflux overlap summary, similar to our existing influx overlap.
• Count total influx/efflux reactions into each microbe in influx/efflux overlap tables. Currently, Nana has been counting the total influx/efflux to each microbe from the trans table.
• Review/edit methods section.

Re run prior analysis used for poster, and also integrate Cindy's analyses (depends on #4)

This should probably be done after #11 is implemented and use FunFlow to cache results.

Currently quite slow; could be improved if made parallel (importantly steadycom/gurobi does not seem to be using more than one cpu in these cases).

The main issue seems to be our memoized function call, which we are also using to save time, takes place in a parfor loop. Replicating the memoized function between worker threads is likely the ideal way to handle this. If it doesn't work, then likely disabling memoization altogether would still be better than the current situation.

• Example showing how to run a simulation (initDMM -> initCobraToolbox -> etc)
• References to container code

See here

Could try mcc first, the though pipes might be easier.

As for the controlling language, use something with good support for recursion. PureScript native is an option, but Idris native might be better thanks to totality checking. See andyarvanitis/idris-cplusplus#1

Check the below as they are completed (we can discuss which ones are not needed):

• createRich.m
• deadendData.m
• excInfo.m (with changes)
• findExcIDs.m (with changes)
• fluxAnalysis.m
• loadModels.m
• multiModels.m
• otherInfo.m
• printBounds.m
• printFluxes.m
• testRun.m

Let's try this:
https://www.mathworks.com/help/matlab/function-based-unit-tests.html

For now we have a prototype at https://github.com/federatedcloud/DouglasMetabolicModels/blob/master/Tests/testMultiModelSingleton.m that can be converted.

More tests for core functionality should be added.

Tests should be run at least before each code commit.

Make new function from here down:

https://github.com/federatedcloud/DouglasMetabolicModels/blob/master/Utils/makeMultiModel.m#L26

https://github.com/tweag/funflow

Possible hash method for matlab files: 

File-based

1. Write each struct to a file named with a random UUID 
2. Make sure it is written to memory (Haskell library) for improved speed (ramfs)
3. checksum or hash file 
4. delete file

Other Appends:

• git hash of relevant repositories
• whether said repositories are dirty
• if dirty, hash the output of git diff or similar
• nix commit
• other deps or data?
  But how does FunFlow store binary data (i.e. data for abstract types)?

In this commit, I notice changes were added to some scripts. For instance, in ecoli_basicTest.m it looks like a rather large section was added to the script, but it doesn't seem to run for me.

Let's try to understand what happened, or if not, revert the change so that it is working again.