HCS-hierarchical algorithm for simulation of omics datasets

HCS-hierarchical algorithm for simulation of omics datasets reconstructs omic datasets that preserve the correlation structure of original data using hierarchical clustering of variables and principal components of clusters, enabling generation of simulated data with implanted ground truth for testing feature selection and machine learning algorithms.

Key Features:

• Hierarchical clustering of variables: Groups variables via hierarchical clustering to reduce the number of parameters while preserving correlation structure.
• Principal components of clusters: Uses principal components derived from clusters to capture and reproduce topological descriptors of the correlation structure.
• Correlation structure preservation: Reconstructs datasets with high-fidelity preservation of the original correlation structure using a reduced parameter set.
• Independent dataset generation: Generates datasets with correlation structures similar to the original data but uncorrelated with the original variables, enabling controlled ground truth implantation.
• Contrast variable generation: Produces contrast variables with desired correlation structures for feature selection evaluation.

Scientific Applications:

• Feature selection algorithm testing: Provides simulated datasets with known ground truth to evaluate and compare feature selection methods on systems mimicking real data structures.
• Machine learning algorithm validation: Enables testing of machine learning methods on omics-like data with controlled correlation structures and implanted ground truth.
• Correlation structure analysis: Supports investigation of topological descriptors of correlation structures in high-dimensional omics data with small sample sizes.

Methodology:

The algorithm performs hierarchical clustering of variables, computes principal components of the resulting clusters, and uses the correlation structure of these principal components to reconstruct datasets with preserved correlation topology.