PARADIGM: PARameters Associated with DynamIcs of Gut Microbiota

This repository contains R scripts to reproduce the five figures in Chi et al., "High resolution analyses of associations between medications, microbiome and mortality in cancer patients".

• R folder contains the key script for PARADIGM algorithm.
• RMD folder contains the scripts to produce Figures 1-5 in the paper.

Data download

Data can be downloaded from Figshare and saved to the data folder (which should be in the same folder as the GitHub repository R and RMD folders):

• https://doi.org/10.6084/m9.figshare.21657806.v1

Data description

1. tblantibiotics_Duke.csv: antibiotic exposure information for patients in the Duke cohort, between day -14 to 14 relative to HCT

• PatientID = deidentified patient IDs
• exposure_name = antibiotic name
• exposure_day_relative_to_hct = date of exposure relative to HCT

2. tblattractor_coefficient_matrix_PARADIGM.csv: raw attractor transition coefficient values from the PARADIGM algorithm, which indicates the associations between drug exposures with cluster attractor transition. This file is the output of paradigm_example_Fig3_bc.RMD script

• Each row corresponds to a given cluster, each column corresponds to a parameter or the intercept
• day.x = time parameter (relative to HCT)

3. tblcounts.csv: Taxonomic classification and ASV counts of samples in the entire study cohort

• count = ASV counts
• count_total = total ASV counts for a particular sampleid
• color = the hex color code for 16S data (used in paradigm_example_Fig2.RMD script)
• color_shotgun = the hex color code for shotgun metagenomic data (used in paradigm_example_Fig2.RMD script)

4. tbldaily_sampling_PARADIGM.csv: dataset of pairs of daily collected samples along with drug exposure records serving as input into the PARADIGM algorithm. This file is the input of paradigm_example_Fig3_bc.RMD script

• PatientID = deidentified patient IDs
• sampleid.x = sample ID of the first sample in a pair of daily collected samples
• sampleid.xy = sample ID of the second sample in a pair of daily collected samples
• n10.x = kmeans cluster assignment of sampleid.x
• n10.y = kmeans cluster assignment of sampleid.y
• day.x = day of collection relative to HCT of sampleid.x
• day.y = day of collection relative to HCT of sampleid.y
• dday = numeric difference between day.x and day.y (should be 1 since we only consider daily collected sample pairs)
• Subsequent columns correspond to a given drug exposure, with a value of FALSE if patients were NOT exposed to the drug on day.x, and a value of TRUE if patients were exposed to the drug on day.x

5. tbldrugs_MSKCC.csv: drug exposure information for patients in the MSKCC cohort, between day -14 to 14 relative to HCT

• PatientID = deidentified patient IDs
• exposure_name = drug name
• exposure_day_relative_to_hct = date of exposure relative to HCT

6. tbleucliean_distance_10clusters_kmeans.csv: a 10-by-10 matrix of pairwise cluster Euclidean distance metrics. Cluster distance is one of the parameters necessary to calculate cluster attractor transition probability.

7. tblgraph_drug_exposure_classification.csv: drug classification for graphing purposes

8. tblmeta_data.csv: raw sequencing file release on NCBI SRA database

9. tblpatient.csv: patient characteristics for the entire study cohort

• PatientID = deidentified patient IDs
• intensity = conditioning intensity
• simplesource = graft source
• disease_simple = underlying disease
• ci = comorbidity index

10. tblresponse_scores_4features_PARADIGM.csv: response scores indicating the association between 62 investigated medications with 4 microbiome features of interest. This file is the output of paradigm_example_Fig3_bc.RMD script

• Each row is a drug exposure, each column is a microbiome feature

11. tblsample.csv: sample characteristics for the entire study cohort

• cluster_assignment = clusters identified by kmeans unsupervised clustering method
• Blautia, Enterococcus, Erysipelatoclostridium = relative abundance of each genus
• tsne1, tsne2 = tSNE coordinates for 16S data (Fig. 1a, c)
• tsne1_shotgun, tsne2_shotgun = tSNE coordinates for shotgun metagenomic data (Fig. 1b)

12. tblself_coefficient_matrix_PARADIGM.csv: raw self transition coefficient values from the PARADIGM algorithm, which indicates the associations between drug exposures with cluster self transition. This file is the output of paradigm_example_Fig3_bc.RMD script

• Each row corresponds to a given cluster, each column corresponds to a parameter or the intercept

13. tblshotgun_MetaPhlAn.csv: taxonomic classification and abundance by MetaPhlAn on shotgun metagenomic samples

14. tblstrain_dynamics.csv: phylogenetic distance of dominant strains across longitudinal sample. This file is the input of pradigm_example_Fig4.RMD script

• PatientID = deidentified patient IDs
• sampleid.x = sample ID of the first sample in a pair of subsequently collected samples
• sampleid.xy = sample ID of the second sample in a pair of subsequently collected samples
• day.x = day of collection relative to HCT of sampleid.x
• day.y = day of collection relative to HCT of sampleid.y
• dday = numeric difference between day.x and day.y
• genus = relative abundance of the genus encompassing species in column "strain_name" in sampleid.x
• species = relative abundance of the species in column "strain_name" in sampleid.x
• phylo_dist = phylogenetic distance of the dominant strains in sampelid.x and in sampleid.y within the species in "strain_name"
• Subsequent columns correspond to a given drug exposure, with a value of FALSE if patients were NOT exposed to the drug between day.x and day.y, and a value of TRUE if patients were exposed to the drug between day.x and day.y.

R dependencies

R dependencies are stored in the renv.lock file; to use this, run renv::restore() in this repository.