seqbias

seqbias corrects protocol-specific sequence biases in high-throughput sequencing reads to improve accuracy of sequence abundance estimates for applications such as RNA-Seq de novo gene annotation and transcript quantification.

Key Features:

• Bias Quantification and Correction: Uses a graphical model to measure and correct sequence biases arising from protocol factors such as polymerase chain reaction amplification and primer affinities.
• Graphical Model Approach: Implements a Bayesian network trained on aligned reads and a reference genome and operates without requiring pre-existing gene annotations.
• Automatic Model Selection: Incorporates automatic model selection to determine model structure or parameters from the data.
• Minimal Risk of Spurious Adjustment: Designed to have negligible effects on unbiased data to avoid introducing false corrections.
• Empirical Validation: Demonstrated across multiple datasets to reduce bias and increase uniformity in sequence abundance measurements.

Scientific Applications:

• RNA-Seq transcript quantification: Improves accuracy of transcript abundance estimates from RNA-Seq data by correcting sequence-specific biases.
• De novo gene annotation: Supports more reliable inference of gene models from sequencing data by reducing protocol-induced bias in read distributions.
• General sequencing bias assessment: Enables evaluation and correction of protocol-specific biases in high-throughput sequencing experiments.

Methodology:

Implements a Bayesian network (graphical model) trained on aligned reads and a reference genome, with automatic model selection to learn and correct protocol-specific sequence biases without relying on gene annotations.