sleuth

sleuth performs differential expression analysis of RNA-seq data by separating biological variance from inferential variance to produce statistically robust transcript- and gene-level expression comparisons.

Key Features:

• Response error linear modeling: Implements a response error linear model to account for measurement error in expression estimates during differential testing.
• Bootstrap-based inferential variance: Incorporates bootstrap-derived uncertainty from transcript quantifications to estimate inferential variance.
• Decoupling of variance components: Distinguishes biological variance from inferential variance to refine differential expression inference.
• kallisto quantification support: Accepts kallisto transcript quantifications, leveraging its pseudo-alignment output and bootstrap replicates.
• Pseudo-alignment reliance: Uses kallisto's pseudo-alignment strategy for rapid transcript quantification feeding into downstream variance-aware analysis.
• Transcript- and gene-level analysis: Enables differential expression comparisons at transcript and aggregated gene levels using quantified abundances and associated uncertainty.
• Computational efficiency: Utilizes fast quantification inputs from kallisto to support scalable analysis of large RNA-seq datasets.

Scientific Applications:

• Differential expression analysis: Detection of differentially expressed transcripts and genes from RNA-seq experiments while accounting for inferential uncertainty.
• Condition and treatment comparisons: Comparison of gene expression changes across experimental conditions or treatments with uncertainty-aware statistics.
• Transcriptomics studies: Application in large-scale transcriptomic investigations where bootstrap-based uncertainty and efficient quantification are required.
• Biological variance characterization: Analysis aimed at separating technical/inferential noise from true biological signal in expression datasets.

Methodology:

Combines kallisto pseudo-alignment quantifications and their bootstrap replicates with a response error linear model to estimate inferential variance and separate it from biological variance for differential expression testing.