RFECS uses a multivariate random forest to capture chromatin signatures at enhancer regions.
Multivariate Random Forest modeling identified seven proteins whose combined altered expression levels effectively segregated affected vs. unaffected twins.
BH: Benjamini and Hochberg; ClustGD: Guided clustering; ClustUN: Unguided clustering; DE: Differentially expressed; GST: Gene set test; MM: Multiple myeloma; MRF: Multivariate random forest; miRNA: microRNA; miRmR: miRNA-mRNA; OR: Odds-ratio; PrEnrich: Percentage of clusters that are enriched.
Our guided clustering method is based on multivariate random forest (MRF) [ 17] and requires two input matrices - a Y × X miRmR map matrix and a Y × T matrix of mRNA expression values, where T denotes the number of conditions (e.g. the number of time points or patient categories).
Dong et al. (2012) compared linear with non-linear models (random forests, multivariate adaptive regression splines) and got similar prediction accuracies.
In contrast to traditional multivariate modeling, the random forest analysis can account for inter-relationships among all potential predictors including highly multilevel unordered categorical covariates in building a tree-based predictive model.
Multivariate linear regression, random forest, and a stepwise selection of linear regression methods were applied: metabolites were chosen if they both were significant in linear regression for every single metabolite, adjusting for BMI and also for the top 30 most important variables in random forest method, in which both the 131 metabolites and BMI were severed as variables.
Note that further methods could have been considered, such as the "traditional" Golub criterion or more sophisticated multivariate approaches, e.g. based on random forests [ 24].
These two models were consecutively constructed using Classification and Regression Tree (CART®), Multivariate Adaptive Regression Splines (MARS®), Random Forests (RF™) and TREENET™ software (Salford Systems, San Diego, CA, USA).
Additional methods of multivariate analysis (partial least squares and Random Forests) also identified maternal third trimester fasting BGL and parity as the most significant predictors of male infant adiposity.
