malacoda

malacoda implements a Bayesian framework to model next-generation sequencing (NGS) counts from massively parallel reporter assays (MPRA) and CRISPR screens to infer functional effects of genetic variants.

Key Features:

• Probabilistic Bayesian Modeling: Implements a fully Bayesian framework to model sequencing data from high-throughput assays and quantify uncertainty.
• Negative Binomial with Gamma Priors: Models sequencing counts using a negative binomial distribution with gamma priors to address overdispersion and sequencing depth dependence.
• Empirical Prior Estimation: Estimates prior distributions empirically from high-throughput assay data.
• Integration of External Annotations: Incorporates external annotations such as ENCODE and DeepSea to inform and refine prior distributions.
• Comprehensive Statistical Framework: Models experimental and technical structure of assays rather than relying on simple transformations of count-level data.
• Quality Control and Utility Functions: Provides quality-control routines and automated barcode counting plus visualization functions for data inspection.

Scientific Applications:

• MPRA variant characterization: Measures transcriptional shifts induced by thousands of genetic variants in MPRA experiments to functionally characterize variants.
• CRISPR screen analysis: Analyzes CRISPR screens using sequencing counts to infer functional effects of perturbations.
• Integrative variant interpretation: Combines assay data with external annotations to refine inference of variant function and link genomic variation to phenotypic outcomes.
• Experimental validation support: Produces candidate variant effect estimates that can be validated with experimental assays such as luciferase reporter assays.

Methodology:

Uses Bayesian statistics with a negative binomial likelihood and gamma priors to model sequencing counts, performs empirical prior estimation from high-throughput data, accounts for input library preparation and sequencing depth, and optionally incorporates external annotations such as ENCODE and DeepSea.