pmm

pmm implements a mixed-model framework for joint statistical analysis of parallel high-throughput RNA interference (RNAi) screening experiments to increase power for hit detection and enable cross-condition comparisons.

Key Features:

• Mixed-model framework: Jointly models multiple RNAi screens to increase sensitivity for hit detection and support cross-condition comparisons.
• R/Bioconductor implementation: Provided as an R package distributed via Bioconductor.
• Weighted analyses: Incorporates screen-level quality metrics and reagent-specific information such as siRNA efficiency scores as model weights.
• Off-target correction: Implements improved procedures for off-target correction in RNAi screens, including settings with limited replication.
• Ranking thresholds: Applies ranking thresholds informed by follow-up screen performance.
• MicroRNA deconvolution: Provides a deconvolution strategy for microRNA mimic screens to enrich the feature space with additional informative measurements.
• Cell-level phenotype integration: Integrates cell-level phenotypes and applies machine-learning approaches such as Random Forests to extract signal absent from aggregate readouts.
• Network inference via off-targets: Uses known off-target relationships among siRNAs as simultaneous perturbations to estimate the inverse covariance matrix of true gene effects for partial correlation–based network inference.
• Simulation benchmarking: Includes simulation studies demonstrating improved accuracy and stability in reconstructing gene–gene dependencies relative to competing approaches.

Scientific Applications:

• Hit detection in RNAi screens: Improved sensitivity for identifying functionally relevant genes, including genes involved in viral or bacterial entry mechanisms.
• Cross-condition comparisons: Comparative analysis of multiple parallel experimental conditions to detect condition-specific and shared effects.
• MicroRNA mimic screen analysis: Deconvolution to expand informative features for microRNA mimic screens.
• Gene–gene network reconstruction: Partial correlation–based inference of gene–gene dependencies using off-target-informed inverse covariance estimation.
• Analysis with limited replication: Procedures tailored to improve robustness of inference in screens with few replicates.
• Extraction of cell-level signals: Use of Random Forests on cell-level phenotypes to detect signals not captured by aggregate readouts.

Methodology:

Methods explicitly include a mixed-model framework for joint analysis of multiple RNAi screens; weighted analyses using screen-level and reagent-specific weights (e.g., siRNA efficiency); improved off-target correction; ranking thresholds informed by follow-up screens; a deconvolution strategy for microRNA mimic screens; integration of cell-level phenotypes via Random Forests; use of known siRNA off-target relationships as simultaneous perturbations to estimate the inverse covariance matrix and enable partial correlation–based network inference; and simulation-based benchmarking.