mLASSO-Hum

mLASSO-Hum predicts protein subcellular localization and identifies explanatory Gene Ontology (GO) terms using a LASSO-based multi-label classification approach.

Key Features:

• Interpretable Predictions: Leverages a LASSO-based approach to produce sparse, interpretable models that highlight GO terms important for protein subcellular localization.
• Multi-Label Classification: Handles multi-label classification to assign multiple potential subcellular localizations to a single protein.
• Sparse Solutions: Employs one-vs-rest LASSO-based classifiers to reduce feature dimensionality and mitigate overfitting in high-dimensional feature vectors.
• Significant GO Terms Identification: From over 8,000 GO terms it identifies 87 essential terms implicated in subcellular localization, providing explanatory features.
• Utilization of Hierarchical Information: Incorporates hierarchical information derived from the depth distance of GO terms to enhance prediction accuracy.
• Comprehensive GO Term Utilization: Integrates GO terms from the cellular component category and other GO categories, contributing to superior performance relative to existing state-of-the-art predictors.

Scientific Applications:

• Disease Mechanism Elucidation: Accurate subcellular localization predictions assist in elucidating mechanisms underlying human diseases.
• Research and Development: Provides interpretable GO-term-based predictions to support proteomics research, experimental validation, and hypothesis generation.

Methodology:

Uses a LASSO-based multi-label classification framework with one-vs-rest classifiers to select sparse, interpretable GO-term features and integrates GO hierarchical information via depth distance.