microage

Scripts for the publication Global analysis of infant gut microbiota revealed distinctive maturation dynamics across lifestyles

File description:

examples for dada2-scripts for studies consisting of one cohort: dada_2_single_cohort.R or multiple cohorts: dada_2_multiple_cohorts.R, both require dada_2_functions.R aggregate all studies and perform rarefaction filtering with merge_all_simple.R alpha and beta diversity analysis, prevalence analyses done with pcoas.R, requires pcoa_functions.R main models are trained with: all_regression_nested_cv.R cross-application of lifestyle specific models with: lifestyle_regression.R, requires lifestyle_regression_functions.R training models on downsampled lifestyle groups with matching age distributions with: lifestyle_regression_downsample.R
all analyses including SAM and preterm infants done with: sick_children.R all random forest scripts need regression_functions.R and alt_models.R (a modified version of ranger for caret to vary selected hyperparameters) scripts for generating all figures in the paper: paper_figures.R and paper_supplement_figures.R

Application of age index

Generate count table following the example dada2-Rscript using the SILVA database version 138

# load functions and model
source(regression_functions.R)
load("../data/final_no_ls_genus.RData") # change here to load different models

# Formatting input data from a phyloseq object
filtered_cdf <- create_caret_df(ps_filtered, transformation = "compositional", only_multiple_samples = F,
                filter_features = names(final_model_genus$finalModel$variable.importance),
                additional_cols = c("Observed", "Shannon"))

# predict microbiome age
predicted_age = predict(final_model_genus_no_ls$rf1, newdata = filtered_cdf$features)

Get SHAP values for new data

shap_res_combined <- get_shap_long(final_model_genus_no_ls$rf1,
                                   test_data = filtered_cdf$features
                                   features = intersect(colnames(filtered_cdf_combined_sub$features), names(final_model_genus$finalModel$variable.importance)),
                                   normalize_ab_values = F)

Comments

I didn't upload all models, further models can either be trained with all_regression_nested_cv.R, or contact me, then I can share them Some intermediate files are too large to be uploaded here, but can be shared on request