SMuRFS

SMuRFS performs sequential multivariate feature selection and inference using random forest (RF) methodologies to identify statistically significant predictors in high-dimensional genomic datasets.

Key Features:

• Sequential Feature Selection: Implements a sequential algorithm that integrates conditional inference trees within the RF framework to remove features based on formal hypothesis testing.
• Conditional Inference Framework: Employs a coherent conditional inference framework that harmonizes variable selection and prediction for inferentially justified selection.
• Random Forest Integration: Leverages random forest (RF) methodologies for both feature importance assessment and downstream predictive modeling.
• High-Dimensional Data Handling: Designed to manage scenarios with numbers of features that often exceed sample sizes, as encountered in genomic studies.
• Predictive Modeling: Constructs predictive RF models using the subset of features retained by the sequential selection process.
• Statistical Significance in Multivariate Contexts: Addresses the identification of statistically significant features in multivariate settings where traditional RF approaches lack formal inferential guarantees.

Scientific Applications:

• Simulation Studies: Demonstrated robust performance in simulation studies assessing feature selection and inference properties.
• Genomics of Drug Sensitivity for Cancer: Applied to the "Genomics of Drug Sensitivity for Cancer" dataset to identify genetic characteristics that influence drug sensitivities, with selected predictors undergoing biological validation.

Methodology:

Uses a sequential algorithm integrating conditional inference trees within a random forest framework, systematically deletes features via explicit hypothesis testing, and builds predictive RF models from the retained predictors under a coherent conditional inference framework.