DeltaNet

DeltaNet predicts genes directly perturbed or targeted by drugs from gene expression data to support drug-target identification.

Key Features:

• Direct Inference: Bypasses separate inference of gene regulatory networks (GRNs) and infers perturbation targets directly from expression data.
• Mathematical Formulation: Frames the problem as an underdetermined linear regression to identify direct gene perturbations.
• Regularization Methods: Implements least angle regression (DeltaNet-LAR) and LASSO regularization (DeltaNet-LASSO).
• Scalability and Robustness: Uses regularization approaches chosen for efficient handling of large datasets while maintaining robustness.
• Parameter Tuning: Requires minimal parameter tuning, particularly with the DeltaNet-LAR variant.

Scientific Applications:

• Drug target identification: Predicts genes directly perturbed or targeted by drugs from gene expression profiles for use in drug-discovery studies.
• Cross-organism evaluation: Applied to predict gene targets in Escherichia coli, yeast, fruit fly, and human cells.
• Benchmarking: Produces predictions reported to be more accurate than mode of action by network identification (MNI) and sparse simultaneous equation model (SSEM) in tested datasets.

Methodology:

Solves an underdetermined linear regression problem using least angle regression (DeltaNet-LAR) or LASSO regularization (DeltaNet-LASSO) while bypassing separate GRN inference.