netboost

Netboost implements a three-step dimension reduction technique that combines boosting-based filtering with network topology measures to identify weak but cumulative feature effects in high-dimensional omics datasets.

Key Features:

• Boosting-based network filtering: Combines a boosting-based filter with the topological overlap measure to identify essential edges of the network.
• Sparse hierarchical clustering: Applies sparse hierarchical clustering on selected edges to identify modules, with bundled MC-UPGMA clustering package by Yaniv Loewenstein.
• Module aggregation: Aggregates module information by the first principal components for dimension reduction.
• Integration with downstream predictors: Compatible with CoxBoost for survival prediction and Random Forest for classification.

Scientific Applications:

• Acute myeloid leukemia survival prediction: Combined with CoxBoost for survival prediction using DNA methylation and gene expression data from AML patients, identifying a signature related to chromatin modifying enzymes replicated in the AMLSG 12-09 study.
• Huntington's disease classification: Combined with Random Forest classification to improve disease classification error in RNA-sequencing data of Huntington's disease mice.
• High-dimensional omics hypothesis generation: Enables dimension reduction and hypothesis generation in high-dimensional omics applications where state-of-the-art selection methods fail to identify weak but cumulative effects.

Methodology:

Netboost employs a three-step approach: first, a boosting-based filter combined with the topological overlap measure identifies essential network edges; second, sparse hierarchical clustering identifies modules from the selected edges; third, module information is aggregated by the first principal components.