GFAsparse

GFAsparse applies Bayesian Group Factor Analysis with element-wise sparsity-inducing priors to identify shared drug response components that link chemical descriptors, including 3D structural descriptors, to genome-wide gene expression across multiple cancer cell lines.

Key Features:

• Multi-Set Analysis Capability: Handles integration of a large number of chemical descriptors against genome-wide gene expression responses across multiple cancer cell lines.
• Bayesian Group Factor Analysis with Element-wise Priors: Implements BGFA with element-wise sparsity-inducing priors to infer sparse, group-level latent factors.
• Identification of Drug Response Components: Extends Group Factor Analysis to extract drug response components linking structural descriptors to gene expression, identifying 11 components across three cancer cell lines in reported analyses.
• Advanced 3D Structural Descriptors: Incorporates advanced 3D structural descriptors to refine associations between chemical features and genomic responses, enabling detection of relationships such as between 15-delta prostaglandin J2 and HSP90 inhibitors.
• Quantitative Performance on Connectivity Map: Demonstrates quantitative improvement over earlier methods applied to the Connectivity Map (CMap) database.

Scientific Applications:

• Pharmacogenomics and Drug-Gene Interaction Discovery: Links drug chemical features to genome-wide expression changes to uncover on-target and off-target effects across cancer cell lines.
• Cross-Cell-Line Comparative Response Analysis: Identifies shared and cell-line-specific response components across multiple cancer models.
• Hypothesis Generation for Disease-Specific Effects: Reveals disease- or lineage-specific responses such as leukemia-specific simvastatin-induced effects resembling corticosteroid responses.

Methodology:

Formulates the task as a search for shared drug response components across multiple cancers and employs Bayesian Group Factor Analysis with element-wise sparsity-inducing priors to extract associations between drug structural descriptors and genome-wide gene expression.