ABOUT.md

• Advanced Config parameters
• Workflow overview
• Package contents

Advanced Config parameters

All advanced config parameters (Advanced) can be used with their default values in mantis_ml/conf/.config. The number of threads and the number of stochastic iterations may be specified by the user during the mantisml run with the -n and -i options respectively.

Workflow overview

mantis-ml follows and Automated Machine Learning (AutoML) approach for feature extraction (relevant to the disease of interest), feature compilation and pre-processing. The processed feature table is then fed to the main algorithm that lies in the core of mantis-ml: a stochastic semi-supervised approach that ranks genes based on their average performance in out-of-bag sets across random balanced datasets from the entire gene pool.

A set of standard classifiers are used as part of the semi-supervised learning task:

• Random Forests (RF)
• Extremely Randomised Trees (Extra Trees; ET)
• Gradien Boosting (GB)
• Extreme Gradient Boosting (XGBoost)
• Support Vector Classifier (SVC)
• Deep Neural Net (DNN)
• Stacking (1st layer: RF, ET, GB, SVC; 2nd layer: DNN)

The parameters from each classifier have been fine-tuned based on grid search using a subset of balanced datasets constructed on the basis of Chronic Kidney Disease example.

Following the semi-supervised learning step of mantis-ml, predictions are overlapped with results from cohort-level rare-variant association studies.

The final consensus results can then been visualised using a set of dimensionality reduction techniques:

• PCA
• t-distributed Stochastic Neighbor Embedding (t-SNE)
• Uniform Manifold Approximation and Projection

Package contents

The mantis-ml-release project contains:

• the mantis-ml package which includes:

  • the mantis-ml main run scripts

    .
    └──mantis_ml
    	└── bin
  

  • example input config files

    .
    └──mantis_ml
    	└── conf
  

  • the mantis-ml modules

      .  
      └──mantis_ml
      	└── modules
    		├── post_processing
    		├── pre_processing
    		│   └── data_compilation
    		├── supervised_learn
    		│   ├── classifiers
    		│   ├── core
    		│   ├── feature_selection
    		│   ├── model_selection
    		│   └── pu_learn
    		├── unsupervised_learn
    		└── validation
  

  • integrated feature data (mantis-ml/data/)

    .
    └──mantis_ml        
    	└── data
    	    ├── adipose_eqtl
    	    ├── ckddb
    	    ├── ensembl
    	    ├── essential_genes_for_mouse
    	    ├── exac-broadinstitute
    	    │   └── cnv
    	    ├── exSNP
    	    ├── genic-intolerance
    	    ├── gnomad
    	    ├── goa
    	    ├── gtex
    	    │   └── RNASeq
    	    ├── gwas_catalog
    	    ├── HPO
    	    ├── human_protein_atlas
    	    ├── in_web
    	    ├── mgi
    	    ├── msigdb
    	    │   └── tables_per_gene_set
    	    ├── neph_qtl
    	    ├── omim
    	    ├── platelets_eqtl
    	    └── rvis_plosgen_2013
  

• output structure examples (out/)

   .
   └── out
	└── CKD-production
		├── data
		│   └── compiled_feature_tables
		├── feature_selection
		│   └── boruta
		│       ├── out
		│       └── tmp
		├── figs
		│   ├── benchmarking
		│   ├── boruta
		│   ├── EDA
		│   ├── supervised-learning
		│   │   └── gene_proba_predictions
		│   └── unsupervised-learning
		├── processed-feature-tables
		├── supervised-learning
		│   ├── gene_predictions
		│   ├── gene_proba_predictions
		│   └── ranked_gene_predictions
		└── unsupervised-learning
 

• Additional analyses modules: misc

  • module for overlap of mantis-ml predictions with cohort-level association studies

        .
        └──misc
    	└── overlap-collapsing-analyses
  

  • module for estimation of Boruta feature importance scores across different disease examples

        .
        └──misc
    	└── boruta-post-processing