PSL-Recommender

PSL-Recommender predicts protein subcellular localization using a recommender-system framework to inform studies of protein function and cellular organization.

Key Features:

• Neighborhood Regularized Logistic Matrix Factorization: Employs neighborhood-regularized logistic matrix factorization to incorporate local structural information and builds upon logistic matrix factorization implementations tailored for implicit-feedback data and drug–target interaction predictions.
• Recommender System Framework: Structures prediction as a recommender system using machine learning techniques to infer likely protein localizations within the cell.
• Benchmarking and Performance: Benchmarked on multiple datasets (one human and three animal datasets) and reported improvements up to 31% in F1-mean, 28% in accuracy (ACC), and 47% in average precision (AVG) versus state-of-the-art methods.

Scientific Applications:

• Protein functional annotation: Uses localization predictions to support inference of protein function and subcellular context.
• Protein interaction studies: Provides localization evidence relevant to mapping protein–protein interactions and interaction contexts.
• Cellular, molecular, and systems biology: Supplies localization data applicable to analyses in cellular biology, molecular biology, and systems biology.

Methodology:

Integrates neighborhood regularization with logistic matrix factorization (neighborhood-regularized logistic matrix factorization) within a recommender-system framework, leveraging prior logistic matrix factorization approaches for implicit-feedback and drug–target interaction prediction.