DMLDA-LocLIFT

DMLDA-LocLIFT predicts the subcellular localization (SCL) of multi-label proteins to identify proteins that reside in two or more cellular compartments.

Key Features:

• Multi-Label Learning Approach: Applies a multi-label learning framework to handle proteins annotated to multiple subcellular locations.
• Feature Encoding Techniques: Represents protein sequences using dipeptide composition, grouped weight encoding, pseudo amino acid composition (PAAC), Gene Ontology (GO) annotations, and pseudo position specific scoring matrix (PSM).
• Dimensionality Reduction via DMLDA: Uses direct multi-label linear discriminant analysis (DMLDA) to reduce the dimensionality of fused feature vectors.
• Label-Specific Features Classifier (LIFT): Classifies refined feature vectors with a Label-specific FeaTures classifier (LIFT) for multi-label localization prediction.

Scientific Applications:

• Multi-label SCL prediction: Predicts subcellular localization for proteins with multiple cellular localizations across taxa.
• Drug discovery and target identification: Provides localization information relevant to protein function to support drug target identification and novel drug design.
• Performance and validation: Validated by jackknife tests on Gram-negative bacteria, Gram-positive bacteria, and plant protein datasets with accuracies of 98.60%, 99.60%, and 97.90%, respectively.

Methodology:

Encode protein sequences with dipeptide composition, grouped weight encoding, PAAC, GO annotations, and pseudo PSM; fuse feature vectors and reduce dimensionality with DMLDA; classify with the LIFT classifier; evaluate performance using jackknife tests.