ProteinLasso

ProteinLasso infers proteins from peptide observations by formulating protein inference as a constrained Lasso regression that incorporates peptide detectability to resolve peptide degeneracy and one-hit wonders.

Key Features:

• Constrained Lasso formulation: Models protein inference as a constrained Lasso regression that leverages peptide detectability to select proteins explaining observed peptides.
• Coordinate descent solver: Uses an efficient coordinate descent procedure to solve the constrained Lasso optimization problem with rapid convergence.
• Ensemble sparsity selection: Employs an ensemble learning strategy to address selection of the Lasso sparsity parameter across datasets.
• Performance and stability: Demonstrated superior identification accuracy, running efficiency, and stability under varying parameter specifications across three experimental datasets.

Scientific Applications:

• Mass spectrometry-based proteomics: Protein identification from peptide-level observations in mass spectrometry proteomic studies.
• Resolving ambiguous peptide mappings: Handling peptide degeneracy and one-hit wonders to improve accuracy of inferred protein lists.

Methodology:

Input: peptide observations from proteomics experiments; Model formulation: constrained Lasso regression based on peptide detectability; Algorithm execution: efficient coordinate descent procedure combined with ensemble learning for sparsity parameter optimization; Output: a subset of candidate proteins that best explain observed peptides, validated on three datasets.