Exploring Heterogeneity in Meta-Analysis using Random Forests

Conduct random forests-based meta-analysis, obtain partial dependence plots for metaforest and classic meta-analyses, and cross-validate and tune metaforest- and classic meta-analyses in conjunction with the caret package. A requirement of classic meta-analysis is that the studies being aggregated are conceptually similar, and ideally, close replications. However, in many fields, there is substantial heterogeneity between studies on the same topic. Classic meta-analysis lacks the power to assess more than a handful of univariate moderators. MetaForest, by contrast, has substantial power to explore heterogeneity in meta-analysis. It can identify important moderators from a larger set of potential candidates (Van Lissa, 2020). This is an appealing quality, because many meta-analyses have small sample sizes. Moreover, MetaForest yields a measure of variable importance which can be used to identify important moderators, and offers partial prediction plots to explore the shape of the marginal relationship between moderators and effect size.

Random Forest Prediction Decomposition and Feature Importance Measure

An R re-implementation of the 'treeinterpreter' package on PyPI < https://pypi.org/project/treeinterpreter/>. Each prediction can be decomposed as 'prediction = bias + feature_1_contribution + ... + feature_n_contribution'. This decomposition is then used to calculate the Mean Decrease Impurity (MDI) and Mean Decrease Impurity using out-of-bag samples (MDI-oob) feature importance measures based on the work of Li et al. (2019) .

Integrative Random Forest for Gene Regulatory Network Inference

Provides a flexible integrative algorithm that allows information from prior data, such as protein protein interactions and gene knock-down, to be jointly considered for gene regulatory network inference.

Block Forests: Random Forests for Blocks of Clinical and Omics Covariate Data

A random forest variant 'block forest' ('BlockForest') tailored to the prediction of binary, survival and continuous outcomes using block-structured covariate data, for example, clinical covariates plus measurements of a certain omics data type or multi-omics data, that is, data for which measurements of different types of omics data and/or clinical data for each patient exist. Examples of different omics data types include gene expression measurements, mutation data and copy number variation measurements. Block forest are presented in Hornung & Wright (2019). The package includes four other random forest variants for multi-omics data: 'RandomBlock', 'BlockVarSel', 'VarProb', and 'SplitWeights'. These were also considered in Hornung & Wright (2019), but performed worse than block forest in their comparison study based on 20 real multi-omics data sets. Therefore, we recommend to use block forest ('BlockForest') in applications. The other random forest variants can, however, be consulted for academic purposes, for example, in the context of further methodological developments. Reference: Hornung, R. & Wright, M. N. (2019) Block Forests: random forests for blocks of clinical and omics covariate data. BMC Bioinformatics 20:358. .

Clustered Random Forests for Optimal Prediction and Inference of Clustered Data

A clustered random forest algorithm for fitting random forests for data of independent clusters, that exhibit within cluster dependence. Details of the method can be found in Young and Buehlmann (2025) .

Random Forests, Linear Trees, and Gradient Boosting for Inference and Interpretability

Provides fast implementations of Random Forests, Gradient Boosting, and Linear Random Forests, with an emphasis on inference and interpretability. Additionally contains methods for variable importance, out-of-bag prediction, regression monotonicity, and several methods for missing data imputation.

Oblique Random Forests for Right-Censored Time-to-Event Data

Oblique random survival forests incorporate linear combinations of input variables into random survival forests (Ishwaran, 2008 ). Regularized Cox proportional hazard models (Simon, 2016 ) are used to identify optimal linear combinations of input variables.

Sequential Permutation Testing of Random Forest Variable Importance Measures

Sequential permutation testing for statistical significance of predictors in random forests and other prediction methods. The main function of the package is rfvimptest(), which allows to test for the statistical significance of predictors in random forests using different (sequential) permutation test strategies [1]. The advantage of sequential over conventional permutation tests is that they are computationally considerably less intensive, as the sequential procedure is stopped as soon as there is sufficient evidence for either the null or the alternative hypothesis. Reference: [1] Hapfelmeier, A., Hornung, R. & Haller, B. (2023) Efficient permutation testing of variable importance measures by the example of random forests. Computational Statistics & Data Analysis 181:107689, .

Integrated Prediction using Uni-Variate and Multivariate Random Forests

An implementation of a framework for drug sensitivity prediction from various genetic characterizations using ensemble approaches. Random Forests or Multivariate Random Forest predictive models can be generated from each genetic characterization that are then combined using a Least Square Regression approach. It also provides options for the use of different error estimation approaches of Leave-one-out, Bootstrap, N-fold cross validation and 0.632+Bootstrap along with generation of prediction confidence interval using Jackknife-after-Bootstrap approach.

Optimising Random Forest Stability by Determining the Optimal Number of Trees

Calculating the stability of random forest with certain numbers of trees. The non-linear relationship between stability and numbers of trees is described using a logistic regression model and used to estimate the optimal number of trees.