Tree-based differential testing using inferential uncertainty for RNA-seq

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Differential expression analysis of RNA-sequencing data is critical for biological insights.
• Transcript abundance estimation uncertainties can lead to false positives or reduced statistical power.
• Existing methods often struggle to effectively incorporate this uncertainty into differential testing.

Purpose of the Study:

• To introduce mehenDi, a novel method for differential transcript expression analysis.
• To leverage a hierarchical tree structure (TreeTerminus) to manage uncertainty in transcript abundance.
• To identify differentially expressed transcripts and transcript groups, including those at inner nodes of the tree.

Main Methods:

• Utilized the TreeTerminus hierarchical structure to represent transcript relationships and uncertainty.
• Developed mehenDi for data-driven differential testing on the tree structure.
• Selected nodes (transcripts or inner nodes) to maximize signal while controlling for abundance uncertainty.
• Compared mehenDi against existing tree-based and uncertainty-aware differential expression methods.

Main Results:

• mehenDi successfully identified differentially expressed inner nodes, revealing signals missed by transcript-only analysis.
• The method demonstrated robust performance on both simulated and experimental RNA-seq datasets.
• mehenDi effectively balances the detection of differential expression with the control of uncertainty.

Conclusions:

• mehenDi offers an advanced approach to differential expression analysis by incorporating transcriptomic hierarchy and uncertainty.
• The method enhances the discovery of biologically relevant differential expression signals.
• mehenDi provides a powerful tool for transcriptomic data interpretation.