Machine learning classification of microbiome data

This workflow is adapted from the Snakemake pipeline of mikropml developed by the Schloss lab.

For more details on these tools, see the Snakemake tutorial and read the mikropml docs.

The Workflow

The Snakefile contains rules which define the output files we want and how to make them. Snakemake automatically builds a directed acyclic graph (DAG) of jobs to figure out the dependencies of each of the rules and what order to run them in. This workflow preprocesses the example dataset, calls mikropml::run_ml() for each seed and ML method set in the config file, and combines the results files.

Installation

1. Install conda and snakemake

2. Clone repository

git clone https://github.com/alexmsalmeida/ml-microbiome.git

How to run

1. Edit the configuration file config/config.yml.

   • dataset: path to the input csv file. Rows as samples and columns as features (e.g., species or genes) with an additional column for the outcome variable per sample.

   • outcome_colname: column name of the outcomes for the dataset.

   • groups_colname: column name of sample grouping (e.g., batch). Use the same identifier for all samples if no grouping required.

   • ml_methods: list of machine learning methods to use. Must be supported by mikropml. Options are:

     • glmnet: linear, logistic, or multiclass regression
     • rf: random forest
     • rpart2: decision tree
     • svmRadial: support vector machine
     • xgbTree: xgboost

   • ncores: the number of cores to use for preprocessing and for each mikropml::run_ml() call. Do not exceed the number of cores you have available.

   • nseeds: the number of different random seeds to use for training models with mikropml::run_ml().

2. (option 1) Run the pipeline locally (adjust -j based on the number of available cores)

snakemake --use-conda -k -j 4

2. (option 2) Run the pipeline on a cluster (e.g., SLURM)

snakemake --use-conda -k -j 100 --profile config/slurm --latency-wait 120

3. View the results in results/performance_results.csv.

By default, the pipeline will not estimate feature importance as this is very time consuming. However, a separate script to perform this analysis can be found in code/get_ml-features.R.