Detecting differential growth of microbial populations with Gaussian process regression.

Peter Tonner, Cynthia Darnell, Barbara E Engelhardt, Amy Schmid

Code repository corresponding to paper "Detecting differential growth of microbial populations with Gaussian process regression."

Setup

Requirements

• git (or download repository as tarball)
• python (virtualenv recommended)

Download this repository

git clone https://github.com/SirRujak/gp_growth_phenotype.git

Setup a local python environment (optional)

virtualenv .

source bin/activate

Install requirements

pip install -r requirements.txt

Examples

For a basic example that does not use jupyter, look at example.py in the notebooks folder.

Input to the B-GREAT method should come from two dataframes, data (n x p) and meta (p x k) where n, p, and k are:

• n: number of time-points
• p: number of time-course measurements (replicates)
• k: number of covariates to be tested, e.g. strain, condition.

The index of data is assumed to be the time column. Each row of meta should correspond to a single column of data, e.g. the number of columns in data should equal the number of rows in meta.

import pandas as pd
data = pd.read_csv("path/to/data.csv",index_col=0)
meta = pd.read_csv("path/to/meta.csv")

assert data.shape[1] == meta.shape[0]

These two variables should then be given to bgreat for use in analysis.

import bgreat
bgreat.setGlobals(_data=data,_meta=meta)

Testing strain effects

For a single effect on growth, a single column in meta will be used for testing differential growth. meta should include a column called strain that defines the strain being growth in each column of data. A specific value in the strain column should correspond to the 'parent' or 'control' strain, and is given to bgreat through the setGlobals function. Additionally, a column should be added to meta called "strain-regression" which represents the column used in building the GP regression model. It should contain 1's and 0's corresponding to a non-parent or parent strain sample, respectively.

parent = 'my-parent'
meta['strain-regression'] = (meta.strain!=parent).astype(int)
bgreat.setGlobals(_parent=parent,_control=control)

A list of mutants can then be given to calculate the Bayes factor score, with permutation testing and FDR calculation.

mutants = ['m1','m2',...]
results = testMutants(mutants)

Testing interactions

Interaction testing requires an additional variable, the control condition, as well as two additional columns in meta - "condition" and "interaction"

control = 'my-control'
meta['interaction'] = ((meta.Condition!=control) & (meta.strain!=parent)).astype(int)
meta['condition'] = (meta.Condition!=control).astype(int)
bgreat.setGlobals(_control=control,_meta=meta)

results = bgreat.testMutantCondition(mutants)