SysSynBio / PheArrMe

Phenotype array guided GEM gap-filling

PheArrMe

Description

PheArrMe (pronounced: "fear me") automates the workflow from phenotype array data to draft genome-scale metabolic network reconstruction. The script takes raw absorbance time-series data from a platereader, generates a list of sole carbon sources for each strain and sets everything up to run CarveMe to generate draft GEMs for each of your collection of strains. Finally, it generates a summary table for each gapfilled model.

Requirements

• CPLEX Studio (Instructions below)
• A conda environment (called PheArrMe) is included environment.yml. Create this environment conda env create -f environment.yml and then switch to the BiologMe environment conda activate PheArrMe
• In my experience, the pip packages don't always install, so you may need to do those manually. They are:
  • carveme
  • biopython
  • cplex (see below)
  • cobra

Note: if you have a mac with Apple Silicon, you can play around with using ARM-64 architecture in various places (CPLEX Studio, most packages on conda) to take advantage of the chip you paid for, but it may cause issues with (at least) bioconda and possibly CarveMe. I recommend just installing everything with the OSX-64 architecture for now. Maybe in a year's time things will play nicer together.

Inputs

• Protein multifasta (.faa) files for each strain in data/genomes/faa/. I've included a script to convert from GenBank format (.gbk), which can be deposited in data/genomes/gbk/ link to dir.
• One of the following:
  • A tab-separated list of sole carbon sources for each strain in data/phearr/soleC_lists/ link to dir. Substrate names should match the biolog ID's. The filename should match the genome filename, but with extension .tsv.
  • A a tab-separated biolog table. The filename format should be <genome>_PM#.tsv, where genome is the filename of genome for the corresponding strain, and # is the plate number (e.g., _PM2.tsv. The format and column headers can be found here.
    • WARNING: If you're exporting a tsv from excel, make sure that it doesn't copy over double-quotes in the column headers. This is weird excel behavior that is difficult to detect and will result in omission of the corresponding substrates from gapfilling.

Libraries

• A base medium composition data/biolog/baseMedium.tsv link to file. A table of all components of the media without a carbon source. This is used to build the media tables. You are encouraged to take a look at this composition and edit as necessary (make sure to use BiGG ID's for metabolites - search for them here).
• A map from Biolog substrate ID's to BiGG ID's data/phearr/mapping_Biolog_BiGG.tsv link to file

Outputs

• Optional: protein sequences in .faa files data/genomes/faa/ link to dir either deposited by the user or generated with downloadNCBI.py link to file.
• Combined media tables formatted for CarveMe in data/biolog/mediaTables/ link to dir
• Gap-filled draft GEM's data/models/ link to dir. These will retain the strain name and will be stored in sbml format (.xml)
• Model summary tables, including model size and gapfilling details data/tables/ link to dir

Modules

• src/downloadNCBI.py link to file Download a .gbk genome file for any organism, saved to data/genomes/gbk/
• src/convert_GBKtoFAA.py link to file Convert .gbk format to .faa, saved to data/genomes/faa/
• src/processPheArr.py link to file Determines which substrates are sole carbon sources for each strain from biolog plots.
• src/biologPlots.py link to file Generates plots of the absorbance time series for all substrates, for each plate, for each strain. Substrates identified in processPheArr.py as sole carbon sources are colored red, no-growth substrates are colored blue.
• src/generateMedia.py link to file Create a concatenated table of all media, saved to data/biolog/mediaTables/
• src/runCarveBatch.py link to file Execute CarveMe for each strain, generates a gapfilled draft GEM and saves to data/models/. (planned: also generate a non-gapfilled model)
• src/analyzeGapfilling.py link to file Generates a table summarizing the gapfilled models, including number of genes, reactions, metabolites, gapfilled reactions, and a list of which reactions were added. This is output to data/tables/modelDescriptions.tsvlink to file

Tests

• test/screenSubstrates.py link to file Checks through all Biolog substrates for ones that are problematic for CarveMe gapfilling. Dumps an error log to the test/ directory test/error.log link to file. I've provided a quick one-liner shell script to return a list of problem media ID's from this error log. I then manually added a boolean 'false' to any problem substrates in the mapping file data/biolog/mapping_Biolog_BiGG.tsv

Usage

Currently, PheArrMe is run with python PheArrMe.py link to file, or you can run a a subset of modules in the order given above.

I started a dev branch to implement a CLI and package the code using either setuptools or conda. Planned arguments and options (in no particular order):

• method for determining growth/no-growth
• run PheArrMe with just an NCBI organism ID
• run PheArrMe with no raw phenotype array data, just a list of known sole C substrates
• option to re-run a gapfill if it fails due to a problematic substrate (by eliminating that substrate). Continue until gapfill completes.
• more to come

CPLEX Studio installation guide

1. Get an academic license through here
2. Navigate to 'Data Science' then 'Software' then 'ILOG CPLEX Optimization Studio' then 'Downloads' then 'IBM ILOG CPLEX Optimization Studio V22.1.1 for OSX ARM64' (or whatever version it currently is). Start the download... it will fail because you need to:
3. Download an installer. Do this and then you can launch the installer to actually:
4. Download the right CPLEX (using the ARM architecture again)
5. Follow the instructions. When done:
6. Switch to the PheArrMe environment. Run the python script python /Applications/CPLEX_Studio2211/python/setup.py install (or whatever version you're using).