MDLoc

MDLoc predicts the subcellular locations of proteins, including multiple simultaneous locations, by modeling inter-dependencies among locations with a Bayesian-network-based probabilistic generative model to support functional interpretation.

Key Features:

• Multi-Location Prediction: Predicts multiple subcellular locations per protein to represent proteins that localize to more than one compartment.
• Dependency Capture: Captures inter-dependencies among different subcellular locations rather than treating locations as independent.
• Probabilistic Generative Model: Employs a mixture-model probabilistic generative framework based on Bayesian networks to represent dependencies between features and locations.
• Iterative Parameter Learning: Uses an iterative process for learning model parameters and estimating protein locations.
• DBMLoc Training and Evaluation: Trains and evaluates models on the DBMLoc dataset, which includes information on both single- and multi-localized proteins.

Scientific Applications:

• Proteomics research: Supports large-scale annotation of protein subcellular localization in proteomics studies.
• Protein functional inference: Aids in inferring protein function by providing location-based context.
• Cellular process analysis: Enables study of cellular processes and localization-dependent mechanisms.
• Translational research: Provides localization information relevant to drug discovery, disease diagnosis, and therapeutic strategy development.

Methodology:

Uses a mixture-model probabilistic generative model based on Bayesian networks with an iterative procedure for learning model parameters and estimating protein locations; models were trained and evaluated on the DBMLoc dataset containing single- and multi-localized proteins.