melanome-ICI-FMPC-analysis

This repository contains supplemental data and analysis source code for Taxonomy-free fecal microbial predictors of immune checkpoint inhibitor benefit and adverse events in melanoma [ref].

Repo structure

• data/: Contains clinical and metagenomic summary data used for analysis, which were published as supplementary data files in the original paper
• src/: Contains source code for major reported analyses
• results/: Scratch directory that src/ writes to if code is rerun

Data files

The following files are included in the data/ folder.

• supp_dataset_1.* : Clinical information for publicly available cohorts utilized in this study, Pittsburgh and New York (training), Dallas (validation) and Houston (test).
• supp_dataset_2.csv : Clinical information for the RadVax cohort (this study)
• supp_dataset_3.csv : FMPC database providing protein IDs, cluster information and functional description for each FMPC
• supp_dataset_4.* : Normalized FMPC counts for the training and validation cohorts
• supp_dataset_5.csv : AUROCs assigned to each FMPC for predicting immunotherapy response in the training and validation cohorts
• supp_dataset_6.csv : Permutation testing results for the 49 selected FMPCs for prediction of immunotherapy response based on the training and validation cohorts.
• supp_dataset_7.csv : Houston test dataset (for immunotherapy response prediction) normalized FMPC abundances for the 49 FMPCs tested for response prediction
• supp_dataset_8.csv : Immunotherapy response prediction AUROCs across the cohorts for the 49 selected response-predictive FMPCs
• supp_dataset_9.csv : AUROCs predicting irAE for each FMPC in the Pittsburgh cohort
• supp_dataset_10.csv : RadVax prospective testing cohort (for irAE prediction) normalized abundances of 5 tested irAE-predictive FMPCs
• supp_dataset_11.csv : AUROCs for irAE prediction for the 5 selected irAE FMPCs across Pittsburgh and RadVax cohorts, with adjustment for covariates.
• supp_dataset_12.txt : Assembled regions of bacteria encoding the identified 4Fe-4S irAE predictive FMPCs
• supp_dataset_13.csv : HLA typing of patients in the RadVax cohort
• supp_dataset_14.csv : Peptide MHC binding prediction results for the RadVax cohort with paired HLA typing
• supp_dataset_15.txt : 4Fe-4S DNA sequences derived for quantification as a proxy for a qPCR test

This study also uses publicly available metagnomic data from the following studies: PRJNA762360, PPRJNA41981, PRJNA397906, PRJEB22893

Code & scripts

The following scripts are included in the src/ folder. Scripts which require additional resources not provided in this repository (e.g. fasta files) are denoted

Scripts can be run in the order specified

• 01_process_data.py : Creates normalized FMPC count table from BlastX output *requires BlastX output & files containing number of lines in the fasta file for each sample
• 02_compute_aucs.py : Computes AUROCs for normalized count matrices. The script assumes the matrices have been subsetted by cohort
• 03_filter_aucs.R : Extracts signigicant FMPCs for downstream analysis
• 04_train_mrs.R : Trains MRS betas
• 05_validate_mrs.R : Validates MRS model
• 06_multi_FMPC_model_train.py : Training/feature selection for the multi-FMPC model
• 07_multi_FMPC_model_validate.py : Validation & exploratory analysis of the muti-FMPC model
• 08_extract_peptides.py : Extract peptides from SPAdes contigs *expects sample level SPAdes contigs as input

Additional code notes:

The original database used to cluster WP_* proteins into FMPCs can be constructed using the following commands

wget 'ftp://ftp.ncbi.nlm.nih.gov/blast/db/nr.*.tar.gz'
cat nr.*.tar.gz | tar -zxvi -f - -C .
blastdbcmd -db path_to_nr_db/nr -taxids 2 -outfmt %f -out fasta_file_name.(fa or fasta)

# Note: path_to_nr_db/ is where the .tar.gz files are extracted.