XBSeq

XBSeq models RNA-seq read counts by incorporating non-exonic mapped reads as Poisson sequencing noise and modeling true expression with a negative binomial distribution to improve differential gene expression detection.

Key Features:

• Statistical Modeling: Models observed signals as a convolution of true expression (negative binomial) and sequencing noise (Poisson) from non-exonic mapped reads.
• Noise Consideration: Explicitly treats non-exonic mapped reads as sequencing noise to better represent background signal in RNA-seq data.
• Performance Evaluation: Evaluated using simulated expression datasets generated under various conditions with parameters derived from real RNA-seq data and compared against other differential expression algorithms.
• Application to Real Data: Applied to real RNA-seq datasets and compared to DESeq, showing comparable performance at baseline noise and improved accuracy as the proportion of non-exonic mapped reads increased.
• Sensitivity to Biological Variability: Assessed sensitivity by varying biological replicates, differential fold changes, and non-exonic expression levels, maintaining high true positive rates with a slight increase in false discovery rate at very low read counts.

Scientific Applications:

• Genome-wide expression profiling: Improves accuracy of differential gene expression analysis in genome-wide RNA-seq studies by accounting for non-exonic reads.
• Noisy dataset analysis: Enhances detection of differentially expressed genes in RNA-seq datasets with significant non-exonic mapped reads.
• Method benchmarking: Enables comparative evaluation of differential expression algorithms using simulations derived from real RNA-seq data and real-data comparisons.

Methodology:

Models observed counts as a convolution of a negative binomial true-expression signal and Poisson noise from non-exonic mapped reads; simulates datasets using parameters derived from real RNA-seq under varied conditions; compares performance to other differential expression algorithms and applies the model to real RNA-seq datasets; sensitivity assessed by varying replicates, fold changes, and non-exonic expression levels.