IntegratedLearner - Integrated machine learning for multi-omics prediction and classification

Introduction

The repository houses the IntegratedLearner R package for multi-omics prediction and classification. Both binary and continuous outcomes are supported.

IntegratedLearner provides an integrated machine learning framework to 1) consolidate predictions by borrowing information across several longitudinal and cross-sectional omics data layers, 2) decipher the mechanistic role of individual omics features that can potentially lead to new sets of testable hypotheses, and 3) quantify uncertainty of the integration process. Three types of integration paradigms are supported: early, late, and intermediate. The software includes multiple ML models based on the SuperLearner R package as well as several data exploration capabilities and visualization modules in a unified estimation framework.

At the core, the IntegratedLearner late fusion algorithm proceeds by 1) fitting a machine learning algorithm (base_learner) per-layer to predict outcome and 2) combining the layer-wise cross-validated predictions using a meta model (meta_learner) to generate final predictions based on all available data points. As a default choice, we recommend Bayesian additive regression trees (BART) as the base learner (base_learner = 'SL.BART') and non-negative least squares/ rank loss minimization as the meta model algorithm (meta_learner = 'SL.nnls.auc'). 'SL.nnls.auc' fits a non-negative least squares (in case of a continuous outcome) and rank loss minimization (in case of binary outcome) on layer-wise cross-validated predictions to generate the final predictions and quantify per-layer contributions.

In addition, >50 ML algorithms are supported. Note that, all the learners must be named such that they are preceeded by the prefix SL. followed by the name of the learner or the associated package (e.g., SL.randomForest, SL.BART, SL.glmnet, etc.). Please check out the SuperLearner user manual for all available options.

Features

• Supports early, late, and intermediate fusion with one line of code
• Dozens of algorithms: XGBoost, Random Forest, GBM, LASSO, SVM, BART, KNN, Decision Trees, Neural Networks, and more
• Integrates with SuperLearner to support even more options to quickly add custom algorithms to the ensemble
• Visualization using built-in plotting
• Hyperparameter tuning
• Screening algorithms
• Options to add new algorithms or change the default parameters for existing ones
• Nested cross-validation to estimate the performance of the integrated machine learner
• Multicore and multinode parallelization for scalability

Dependencies

IntegratedLearner requires the following R package: devtools (for installation only). Please install it before installing IntegratedLearner , which can be done as follows (execute from within a fresh R session):

install.packages("devtools")
library(devtools)

Installation

Once the dependencies are installed, IntegratedLearner can be loaded using the following command:

devtools::install_github("himelmallick/IntegratedLearner")
library(IntegratedLearner)

Basic Usage

IntegratedLearner(feature_table, sample_metadata, feature_metadata, ...)

Arguments

• feature_table : Data frame representing concatenated multi-omics features with features in rows (rownames) and samples in columns (colnames).
• sample_metadata : Data frame of sample-specific metadata. Must have a column named subjectID describing per-subject unique identifiers. For longitudinal designs, this variable is expected to have non-unique values. Additionally, a column named Y must be present which is the outcome of interest (can be binary or continuous). Row names of sample_metadata must match the column names of feature_table.
• feature_metadata : Data frame containing feature-specific metadata. Must have a column named featureID describing per-feature unique identifiers. Additionally, if multiple omics layers are present, a column named featureType should describe the corresponding source layer (e.g. metagenomics, metabolomics, etc.). Row names must match that of feature_table.
• feature_metadata_valid : Optional feature table from validation set. Must have the exact same structure as feature_table.
• sample_metadata_valid: Optional sample-specific metadata table from independent validation set. Must have the exact same structure as sample_metadata.
• family: A character string representing one of the built-in families. Currently, gaussian() and binomial() are supported.
• folds: Integer. Number of folds for cross-validation. Default is 5.
• base_learner : Character string representing the name of the SL base-learner in stacked generalization and optionally for joint learner (see example). Check out the SL user manual for all available options. Default is 'SL.BART'
• meta_learner: Character string representing the name of the SL meta-learner in stacked generalization (see example). Check out the SL user manual for all available options. Default is 'SL.nnls.auc'
• run_concat: Logical value representing whether a joint (concatenated) model should also be run (see tutorial). Default is TRUE.
• run_stacked: Logical value representing whether a stacked model should also be run (see tutorial). Default is TRUE.
• print_learner: Logical value representing whether a summary of fit should be printed. Default is TRUE.
• verbose: Logical value for printing progress during the computation (helpful for debugging). Default is FALSE.
• ...: Additional arguments for SL tuning parameters.

The IntegratedLearner workflow

Value

• SL_fits: A list of SL prediction results from all individual base learners, the meta learner, and optionally the joint (concatenation) learner.
• model_fits: A list of base_learner objects extracted from SL_fits for all individual base learners, meta learner, and optionally the joint (concatenation) learner.
• X_train_layers: Input feature matrices for individual layers for training data.
• Y_train: Input response vector for training data.
• yhat.train: Predictions for training data from all individual base learners, the meta learner, and optionally the joint (concatenation) learner.
• X_test_layers: Input feature matrices for individual layers for test data. Available if feature_table_valid is provided.
• Y_test: Input response vector for test data.
• weights: Estimated layer weights in the meta model. Available if run_stacked=TRUE and meta_learner='SL.nnls.auc'.
• AUC.train/R2.train: AUC/ R2 metrics calculated on training data using yhat.train and Y_train.
• AUC.test/R2.test: AUC/ R2 metrics calculated on test data using yhat.test and Y_test.
• ...: Additional arguments containing information about inputs.

Getting Started

The package vignette demonstrates how to use the IntegratedLearner workflow to perform a multi-omics prediction and classification task. This vignette can be viewed online here.

Citation

If you use IntegratedLearner in your work, please cite the following:

Mallick H et al. (2024). An Integrated Bayesian Framework for Multi-omics Prediction and Classification. Statistics in Medicine 43(5):983–1002.

Issues

We are happy to troubleshoot any issues with the package. Please contact the maintainer via email or open an issue in the GitHub repository.

Future Release

We are currently in the process of submitting IntegratedLearner to Bioconductor. Likewise, please keep an eye out for a future release of IntegratedLearner as an R/Bioconductor package while this repository remains the development version of the package.