IsoLasso

IsoLasso reconstructs full-length mRNA transcripts and estimates their expression levels from RNA-Seq short reads using a LASSO-based regression framework to produce accurate, parsimonious, and complete transcriptome assemblies.

Key Features:

• Maximization of Prediction Accuracy: Estimates expression levels so assembled transcripts closely match observed expression across expressed genomic regions.
• Minimization of Interpretation (Parsimony): Applies a parsimony principle to predict the minimal number of transcripts necessary to explain observed expression.
• Maximization of Completeness: Seeks to explain the maximum number of mapped reads or expressed segments within gene models by ensuring containment within predicted transcripts.
• LASSO-based quadratic programming: Implements the Least Absolute Shrinkage and Selection Operator (LASSO) as a multivariate regression with additional constraints formulated in a quadratic programming framework to balance accuracy and interpretability.
• Sensitivity and precision: Demonstrates higher sensitivity and precision on simulated and real RNA-Seq datasets compared to existing tools.

Scientific Applications:

• Transcriptome reconstruction: Reconstruction of full-length mRNA transcripts from short-read RNA-Seq data to produce comprehensive representations of the transcriptome.
• Expression estimation and gene-model interpretation: Estimation of transcript expression levels and parsimonious explanation of mapped reads or expressed segments within gene models for downstream gene expression and regulation studies.

Methodology:

Applies the LASSO (Least Absolute Shrinkage and Selection Operator) multivariate regression and incorporates additional constraints into a quadratic programming formulation to select a parsimonious set of transcripts while fitting observed expression across expressed genomic regions.