Uncovering the link between gut microbiome, highly-processed food consumption and diet quality through bioinformatics methods

This repository contains all scripts and auxiliary data used to analyse metagenomic data from two cohorts, perform differential abundance analyses (DAA) and build all Machine Learning classifiers, as well as regarding the validation of one of these ML models and the figures shown in this MSc Thesis' memory.

Table of contents

• Data directory
• Files for heatmap construction
• R scripts
• Taxonomic assignment

Data directory

This directory contains all files used for the training of all ML classifiers, for the generation of both heatmaps shown in the TFM Thesis and for the plotting of all figures shown in the same document. Other data relative to ML validation and taxonomic assignment are also available here. These files are:

• One directory called importances_HPF. Six files are stored here, where each file contains importances for all 20 taxa identified to be important by its corresponding model and the HPF classification of the study cohort:

  • imp_caret_gbm.csv
  • imp_caret_rf.csv
  • imp_caret_svm.csv
  • imp_mikropml_rf.csv
  • imp_mikropml_svm.csv
  • imp_mikropml_xgbtree.csv

• One directory called importances_HEI. Six files are stored here, where each file contains importances for all 20 taxa identified to be important by its corresponding model and the HEI classification of the study cohort:

  • imp_caret_gbm_hei.csv
  • imp_caret_rf_hei.csv
  • imp_caret_svm_hei.csv
  • imp_mikropml_rf_hei.csv
  • imp_mikropml_svm_hei.csv
  • imp_mikropml_xgbtree_hei.csv

• class_hei_groups.tsv: this file indicates which HEI group each sample from the study cohort belongs to.

• class_hpf_groups.tsv: this file indicates which HPF group each sample from the study cohort belongs to.

• data_clr.tsv: this file is the CLR-transformed relative abundances table relative to the study cohort, without indicating which group each sample pertains to.

• data_clr_hei.csv: this file contains the same information as data_clr.tsv but with an additional final column indicating which HEI group each sample from the study cohort belongs to.

• data_clr_hpf.tsv: this file contains the same information as data_clr.tsv but with an additional final column indicating which HPF group each sample from the study cohort belongs to.

• filtered_table.qza: this is one of the QIIME2 artefacts needed to create phyloseq objects from the study cohort necessary for diversity analyses, DAA and ML classifier construction. If transformed into a visualisation file (.qzv), it informs us about the number of samples taken into account, the number of identified features, etc.

• full_data_population_tables.csv: this file contains all metadata from both cohorts merged in this one .csv table. Variables only available in one of the two cohorts are also shown, with NAs indicating missing data.

• metadata_study_cohort.csv: this .csv table contains all metadata from the study cohort.

• metadata_validation_cohort.csv: this .csv table contains all metadata from the validation cohort.

• metaphlan_abundances_validation_cohort.csv: this .csv table contains the relative abundances table obtained after executing MetaPhlAn4 on the raw reads from the validation cohort.

• rooted-tree.qza: this is one of the QIIME2 artefacts needed to create phyloseq objects from the study cohort necessary for diversity analyses, DAA and ML classifier construction. It is the rooted tree obtained during read processing with QIIME2, MAFFT and FastTree.

• taxonomy.qza: this is one of the QIIME2 artefacts needed to create phyloseq objects from the study cohort necessary for diversity analyses, DAA and ML classifier construction. If transformed into a visualisation file (.qzv), it informs us about the number of samples taken into account, the number of identified features, etc. If transformed into a visualisation file (.qzv), it informs us about the full taxonomy of each identified feature.

• test_data_for_validation.csv: this is a .csv table containing all CLR-transformed abundances for the 200 taxa used for ML training and in relation to the validation cohort. This is the data that will be used as "test data" when asking the most precise ML classifier, for the HEI classification, to predict its HEI classes in the validation procedure.

Files for heatmap construction

The heatmap_files directory contains all data necessary for the construction of heatmaps shown in the MSc Thesis (Figures 4 and 7):

• One directory called additional_files_hpf_classification. Three files are stored here:
  • clr_abundances_for_heatmap.csv: this file contains CLR-transformed relative abundances for the 20 most important taxa according to the most precise ML classifier based on HPF consumption, where each row is a sample and each column is a genus. Samples are ordered so that those belonging to the "Low HPF consumption" group appear first, followed by those from the "High HPF consumption" group.
  • groups_for_heatmap_genuses.csv: this file consists in 20 rows and 2 columns: rows are for each one of the 20 most important taxa according to the best classifier based on this classification of the study cohort, while the first column are these taxa and the second one indicates the HPF consumption group in which they are enriched.
  • groups_for_heatmap.csv: this file indicates which HPF consumption group each sample of the study cohort belongs to. There are therefore as many rows as samples there are in this cohort, while the second column details the group each individual pertains to.
• One directory called additional_files_hei_classification. Three files are stored here:
  • clr_abundances_for_heatmap_hei.csv: this file contains CLR-transformed relative abundances for the 20 most important taxa according to the most precise ML classifier based on HEI, where each row is a sample and each column is a genus. Samples are ordered so that those belonging to the "Good HEI" group appear first, followed by those from the "Poor HEI" group.
  • groups_for_heatmap_genuses_hei.csv: this file consists in 20 rows and 2 columns: rows are for each one of the 20 most important taxa according to the best classifier based on this classification of the study cohort, while the first column are these taxa and the second one indicates the HEI group in which they are enriched.
  • groups_for_heatmap_hei.csv: this file indicates which HEI group each sample of the study cohort belongs to. There are therefore as many rows as samples there are in this cohort, while the second column details the group each individual pertains to.

R scripts

The R_scripts directory contains:

• alpha_and_beta_diversity.R: R script describing all procedures needed to calculate alpha and beta diversity measures and plot several indices for both cohorts and classifications.
• best_ML_model_study_cohort_HPF.R: R script describing how the most precise model for the HPF-based classification was built. Feature importances are calculated as well. This script also includes the code needed to draw the heatmap shown in Figure 4 of this MSc Thesis, relative to the HPF classificafion of the study cohort.
• best_ML_model_study_cohort_HEI.R: R script describing how the most precise model for the HEI-based classification was built. Feature importances are calculated as well. This script also includes the code needed to draw the heatmap shown in Figure 7 of this MSc Thesis, relative to the HEI classificafion of the study cohort.
• boxplots.R: R script indicating in detail how boxplots showing taxa being significantly different in their CLR-transformed relative abundances between HPF- or HEI-based groups were made (Supp. Figs 3, 10 and 11). It is therefore indicated how those individuals with type 2 diabetes family history were highlighted, as well as how the Wilcoxon rank sum test was used to find these differentially abundant taxa between groups.
• daa.R: R script describing how differential abundance analyses were performed for both classifications of the study cohort and the validation one grouped based on their HEI.
• importance_barplots.R: R script describing how Supplementary Figures 2 and 9, consisting in barplots and showing the most important taxa according to each one of our ML classifiers, were made.
• metaphlan_to_R.R: R script indicating how, once the relative abundances table from the validation cohort was obtained after executing MetaPhlAn4, these abundaces were imported into R for the creation of all phyloseq objects necessary for diversity analyses, DAA and ML classifier validation.
• ML_models_HEI.R: R script indicating in detail how all 6 ML models for the HEI classification of the study cohort were built.
• ML_models_HPF.R: R script indicating in detail how all 6 ML models for the HPF classification of the study cohort were built.
• population_tables.R: R script describing how Tables 1-3, which showed some demographic and additional data regarding both cohorts in the MSc Thesis, were created using the gtsummary R package.
• qiime_to_R.R: R script indicating how, once all QIIME2-related procedures for the study cohort were finished, the resulting artefacts were used for the creation of phyloseq objects in R necessary for diversity analyses, DAA and ML classifier construction.
• quantile_graphs.R: R script describing how Supplementary Figures 12 to 15, showing boxplots of some differently abundant taxa vs quantiles made using the validation cohort based on their levels of several glycaemic traits (HOMA-IR, HbA1c (%) and adiponectin) and HEI, were obtained. The identification of these taxa differentially abundant among these quantiles via the Kruskal-Wallis rank sum test is also detailed in this script.
• roc_curves.R: R script detailing how ROC curves for all ML classifiers regarding both classifications of the study cohort were made (Figures 3 and 5, as well as Supplementary Figures 1 and 8).
• validation.R: R script describing how validation of the best ML classifier trained on the study cohort classified based on their HEI was performed using the relative abundances table from the validation cohort, along with the plotting of the ROC curve linked to this validation (Figure 6 of the MSc Thesis).

Taxonomic assignment

The taxonomic_assignment directory contains:

• 16s_data.sh: bash script used for the processing of raw reads using QIIME2 and their taxonomic assignment, until obtaining all necessary files for the generation of phyloseq objects that will be needed for ML classifier construction based on HPF consumption and HEI.
• wgs_data.sh: bash script used for the processing of raw reads using metaWRAP and MetaPhlAn4 and their taxonomic assignment, until obtaining the relative abundances table necessary for the generation of phyloseq objects that will be needed for diversity analyses, DAA and ML validation.