MARSpred

MARSpred predicts the sub-cellular localization of aminoacyl-tRNA synthetases (AARSs) in eukaryotic cells, distinguishing mitochondrial from cytosolic variants to inform studies of mitochondrial protein synthesis.

Key Features:

• SVM-Based Prediction Models: Employs support vector machine (SVM) modules that analyze amino acid composition, dipeptide composition, position-specific scoring matrices (PSSMs), and separate sequence regions (N-terminal, intermediate, C-terminal).
• Split Amino Acid Composition (SAAC) and Selected Attributes (SA-SAAC): Uses SAAC to evaluate distinct sequence regions and refines features via selected attributes (SA-SAAC) to enhance discrimination between mitochondrial and cytosolic AARSs.
• Performance Metrics: Reported performance includes MCC 0.92 and 96% accuracy using SA-SAAC, independent dataset MCC 0.95 and 97.77% accuracy, with module MCCs: amino acid composition 0.82, dipeptide composition 0.73, PSSM 0.78, and SAAC range 0.39–0.86.
• Evaluation Strategy: Models were trained and tested on a non-redundant dataset using fivefold cross-validation to assess reliability and generalizability.

Scientific Applications:

• Mitochondrial biology: Facilitates identification of AARSs that function within mitochondria versus the cytosol to support studies of mitochondrial protein synthesis.
• Functional annotation of AARSs: Provides localization assignments that aid interpretation of enzymatic roles and cellular compartmentalization of AARSs.
• Evolutionary studies: Supports analyses of endosymbiotic gene transfer and evolutionary implications by distinguishing mitochondrial-targeted nuclear-encoded AARSs from cytosolic counterparts.

Methodology:

Developed SVM modules using features from amino acid and dipeptide compositions, PSSMs, region-wise SAAC (N-terminal, intermediate, C-terminal) and selected SA-SAAC attributes, with training/testing on a non-redundant dataset evaluated by fivefold cross-validation.