redox-proteome

Though the term "proteome" is sometimes used to refer to the set of coding sequences in a genome, e.g. as in "reference proteome," here proteome will refer to the relative or absolute levels of expressed proteins.

Dependencies

Code in this project is written entirely in Python, with some analysis in Mathematica. Python dependencies can be installed with pip or conda as below.

Install statistical and numerical python packages

pip install numpy scipy scikit-learn pandas pingouin

plotting utilities

pip install matplotlib seaborn

numerical optimization and metabolic network models

pip install cobra cvxpy 

and various utilities

pip install biopython tqdm lxml

Directories

• data/ contains source and derived data in a directory hierarchy
• munge/ contains scripts that pre-process data for analysis and plotting
• notebooks/ contains scripts and notebooks that perform analysis
• notebooks/linear_opt/ contains code for optimizing the linear form of our model
• mathematica/ contains Mathematica notebooks
• models/ contains files that define models used in code

Performing analyses

To perform analyses needed to generate figures, you will need to first retrieve some data that is too large to host here --- UniProt reference proteomes, GTDB sequences, etc. This is documented in the sections below.

After retrieving this data, you will need to run the scripts in munge/. These are individually documented and perform tasks like merging reference proteomes with expression data, calculating protein $Z_C$ values and computing correlations.

Batch calculation of mean coding sequence $Z_C$ for GTDB representative genomes (munge/calc_genome_nosc_batch.py) is intended to run on a multicore system. It will be very slow on a single computer (was run on 48 cores). As such, I have provided output in data/gtdb/r207/genome_average_nosc.csv.

To perform optimizations of the linearized model used to generate figures 1-3, run notebooks/do_optimization_analyses.py. This should take a few minutes. Simulations of non-linear models are performed in scripts in the akshit_notebooks/ folder.

Generating figures

Figures are all generated from iPython notebooks in the notebooks/ directory. The relevant notebooks have the prefix Fig. Once the relevant pre-processing is done, these should run quickly.

Reference coding sequences

Reference coding sequences live in data/genomes/ and were drawn from UniProt entries for E. coli (UP000000625_83333.xml), yeast (UP000002311_559292) and cyanobacteria (UP000001425_1111708) proteins. Full documentation of the reference sequences used is give in data/genomes/reference_proteomes.csv.The iPython notebook notebooks/00_PreprocessCodingSequences.ipynb extracts these coding sequences and related metadata from UniProt XML files and calculates $Z_C$ values for each coding sequence.

Amino acid properties

Amino acid molecular weights, carbon content and $Z_C$ values are found in data/aa_nosc.csv. $Z_C$ values are drawn from LaRowe & Amend, ISME J 2016 and were checked by manual calculation (TODO). Polar requirement and hydropathy values are drawn from Shenhav & Zeevi, Science 2020, who draw those values from Haig & Hurst J. Mol. Evol. 1991.

Proteomics data

Proteome data was downloaded from the relevant references (as described in the methods section), reformatted and stored in data/proteomes.

GTDB

Data from the Genome Taxonomy Database (GTDB) is drawn from version 207, downloaded from https://gtdb.ecogenomic.org/downloads and stored in data/gtdb. We are working with the $\approx 60,000$ representative genomes rather than the full set of genomes since this limits computational complexity and, further, these genomes are, where possible, drawn from type strains with a higher degree of completeness and better annotation.