This repository hosts code for the in-progress manuscript Bayesian inference of metabolic kinetics from genome-scale multiomics data by Peter C. St. John, Jonathan Strutz, Linda J. Broadbelt, Keith E.J. Tyo, and Yannick J. Bomble, https://doi.org/10.1101/450163.
General code for solving for the steady-state metabolite and flux values as a function of elasticity parameters, enzyme expression, and external metabolite concentrations is found in emll/linlog_model.py. Theano code to perform the regularized linear regression (and integrate this operation into pymc3 models) is found in emll/theano_utils.py.
The notebooks directory contains the main code used to generate figures in the manuscript. wu2004.ipynb contains a simple model of an in vitro pathway, used to compare NUTS and ADVI inference methods. contador.ipynb compares the given methodology to an earlier application of metabolic ensemble modeling. hackett.ipynb demonstrates how the method can scale to near genome-scale models and omics datasets.
A duplicate of the python enviroment I used to perform the calculations should be creatable using anaconda
$ conda env create -f environment.yml
$ source activate idp_new
It uses the intelpython distribution for some faster blas routines, at least on the processors I developed this method on.