uslcount

uslcount identifies and corrects gene expression inaccuracies in unstranded RNA-seq data by predicting mis-quantified genes and refining differential expression estimates.

Key Features:

• Identification of Erroneous Estimates: Leverages a comprehensive stranded RNA-seq dataset encompassing 15 blood cell types to identify genes likely inaccurately quantified in unstranded data, finding approximately 10% of all genes and 2.5% of protein-coding genes exhibit ≥2-fold discrepancies when strand orientation is ignored.
• Machine Learning Model: Employs a machine learning model constructed using parameters from read alignments to predict which genes in an unstranded dataset may have incorrect expression estimates.
• Correction of Differential Expression: Refines differential expression analysis by focusing on reads that span exonic boundaries to mitigate biases introduced by unstranded read datasets.

Scientific Applications:

• Transcriptomic Profiling: Improves the accuracy of gene expression quantification in studies using unstranded RNA-seq data, enhancing transcriptomic analyses across biological samples.
• Differential Expression Analysis: Recovers more accurate differential expression results from unstranded datasets by identifying and correcting genes with strand-related quantification bias.

Methodology:

Uses a stranded RNA-seq reference from 15 blood cell types to identify problematic genes, derives parameters from read alignments for a machine learning model to predict mis-quantified genes, and restricts differential expression to reads spanning exonic boundaries.