NLDMseq

NLDMseq estimates gene and isoform expression from RNA-Seq data by modeling latent isoforms and isoform- and exon-specific read sequencing biases to account for ambiguous read mapping and non-uniform read distribution.

Key Features:

• Latent variable model: Uses latent variables to represent unknown isoforms and the underlying proportions of multiple spliced variants.
• Bias modeling: Explicitly models isoform- and exon-specific read sequencing biases.
• Replicate-based bias identification: Identifies read-distribution biases using replicate information from multiple lanes of a single library run.
• Ambiguous read mapping handling: Accounts for ambiguous mapping of reads to a reference transcriptome caused by alternative splicing.
• Non-uniform read distribution: Addresses positional biases, sequencing artifacts, mappability issues, and other factors that violate uniform read distribution assumptions used by RPKM and Poisson-based models.
• Differential expression support: Applicable to differential expression (DE) detection in downstream transcriptomic analyses.
• Evaluation: Validated on simulation studies and real datasets with reported competitive accuracy compared to existing popular methods.

Scientific Applications:

• Gene and isoform expression estimation: Quantifies gene- and isoform-level expression from RNA-Seq data.
• Differential expression detection: Enables DE analysis at gene and isoform resolution.
• Alternative splicing analysis: Supports quantification in the presence of multiple spliced variants.
• Bias-aware transcriptome quantification: Improves expression estimates when positional biases, sequencing artifacts, or mappability issues are present.

Methodology:

Implements a latent variable model representing unknown isoforms and their proportions, explicitly models isoform- and exon-specific read sequencing biases, and identifies those biases using replicate information from multiple lanes of a single library run.