BRACoD 0.2.9

BRACoD is a method to identify associations between bacteria and physiological variables in Microbiome data

BRACoD: Bayesian Regression Analysis of Compositional Data

Installation

Installation in python:

pip install BRACoD

There is also an R interface, which depends on the python version being installed. There is a helper function that will do it for you, but it might be easier to do it with pip.

devtools::install_github("ajverster/BRACoD/BRACoD.R")

Python Walkthrough

1. Simulate some data and normalize it

   import BRACoD
   sim_counts, sim_y, contributions = BRACoD.simulate_microbiome_counts(BRACoD.df_counts_obesity)
   sim_relab = BRACoD.scale_counts(sim_counts)
   

2. Run BRACoD

   trace = BRACoD.run_bracod(sim_relab, sim_y, n_sample = 1000, n_burn=1000, njobs=4)
   

3. Examine the diagnostics

   BRACoD.convergence_tests(trace, sim_relab)
   

4. Examine the results

   df_results = BRACoD.summarize_trace(trace, sim_counts.columns, 0.3)
   

5. Compare the results to the simulated truth

   bugs_identified = df_results["bugs"].values
   bugs_actual = np.where(contributions != 0)[0]
   
   precision, recall, f1 = BRACoD.score(bugs_identified, bugs_actual)
   print("Precision: {}, Recall: {}, F1: {}".format(precision, recall, f1))
   

6. Try with your real data. We have included some functions to help you threshold and process your data

   df_counts = BRACoD.threshold_count_data(BRACoD.df_counts_obesity)
   df_rel = BRACoD.scale_counts(df_counts)
   df_rel, Y = remove_null(df_rel, BRACoD.df_scfa_obesity["butyric"].values)
   trace = BRACoD.run_bracod(df_rel, Y, n_sample = 1000, n_burn=1000, njobs=4)
   df_results = BRACoD.summarize_trace(trace, sim_counts.columns, 0.3)
   

R Walkthrough

1. Simulate some data and normalize it

   library('BRACoD.R')
   data(obesity)
   r <- simulate_microbiome_counts(df_counts_obesity)
   
   sim_counts <- r[[1]]
   sim_y <- r[[2]]
   contributions <- r[[3]]
   sim_relab <- scale_counts(sim_counts)
   

2. Run BRACoD

   trace <- run_bracod(sim_relab, sim_y, n_sample = 1000, n_burn=1000, njobs=4)
   

3. Examine the diagnostics

   convergence_tests(trace, sim_relab)
   

4. Examine the results

   df_results <- summarize_trace(trace, colnames(sim_counts))
   

5. Compare the results to the simulated truth

   bugs_identified <- df_results$bugs
   bugs_actual <- which(contributions != 0)
   
   r <- score(bugs_identified, bugs_actual)
   
   precision <- r[[1]]
   recall <- r[[2]]
   f1 <- r[[3]]
   
   print(sprintf("Precision: %.2f, Recall: %.2f, F1: %.2f",precision, recall, f1))
   

6. Try with your real data. We have included some functions to help you threshold and process your data

   df_counts_obesity_sub <- threshold_count_data(df_counts_obesity)
   df_rel <- scale_counts(df_counts_obesity_sub)
   r <- remove_null(df_rel, Y)
   df_rel <- r[[1]]
   Y <- r[[2]]
   
   trace <- run_bracod(df_rel, Y, n_sample = 1000, n_burn=1000, njobs=4)
   df_results <- summarize_trace(trace, colnames(df_counts_obesity_sub), 0.3)