casper

casper performs inference of alternative splicing from RNA-sequencing (RNA-seq) data using novel data summaries and a Bayesian non-parametric framework to estimate isoform/path usage and adjust technical biases.

Key Features:

• Novel data summaries: Generates data summaries that capture comprehensive information from RNA-seq datasets, including relationships from paired-end reads, to represent high-dimensional splicing signals.
• Bayesian modeling framework: Implements a Bayesian framework that flexibly estimates technical biases in a non-parametric manner.
• Path counting across exons: Counts read-supported paths across exons rather than relying solely on pairwise exon connections.
• Non-parametric fragment modeling: Estimates fragment size and fragment start distributions non-parametrically.
• Point estimates and uncertainty assessment: Produces efficient point estimates and quantifies uncertainty for individual samples and across datasets.
• Improved estimation accuracy: Demonstrates several-fold improvement in estimation mean square error (MSE) in simulations and substantially higher consistency between replicates in experimental data.
• Adaptability to sequencing technologies: Non-parametric components provide flexibility to adapt to advancements in paired-end sequencing and RNA-seq protocols.

Scientific Applications:

• Alternative splicing quantification: Quantifying alternative splicing and isoform/path usage from RNA-seq data in studies of cellular function and disease mechanisms.
• Reproducibility and replicate analysis: Improving consistency and reproducibility of splicing estimates across biological replicates and experimental datasets.
• Method benchmarking: Benchmarking and method comparison using simulations to evaluate estimation mean square error and replicate consistency.

Methodology:

Computational steps include computing novel RNA-seq data summaries from paired-end reads, counting exon-spanning paths, and performing Bayesian inference with non-parametric estimation of fragment size, fragment start distributions, and technical biases to produce point estimates and uncertainty measures.