PFMpred

PFMpred predicts mitochondrial proteins of the malaria parasite Plasmodium falciparum to support identification of mitochondrial localization and prioritization of proteins for functional and therapeutic studies.

Key Features:

• Prediction target: Predicts mitochondrial versus non-mitochondrial proteins in Plasmodium falciparum.
• Machine learning model: Uses Support Vector Machine (SVM) models trained on a dataset of 175 proteins (40 mitochondrial, 135 non-mitochondrial) with five-fold cross-validation.
• Amino acid composition baseline: Employs simple amino acid composition analysis achieving a Matthews Correlation Coefficient (MCC) of 0.38.
• Split Amino Acid Composition (SAAC): Separately analyzes N-termini, C-termini, and remaining sequence composition, achieving an MCC of 0.73.
• PSSM profile integration: Incorporates Position-Specific Scoring Matrix (PSSM) profiles, achieving an MCC of 0.75 and accuracy of 91.38%.
• Hybrid model: Combines PSSM profiles with SAAC into a hybrid model, yielding an MCC of 0.81 and accuracy of 92%.
• Independent evaluation: Validated on independent datasets and reported to outperform existing methods on mitochondrial protein prediction for P. falciparum.

Scientific Applications:

• Mitochondrial localization annotation: Assigns mitochondrial localization to P. falciparum proteins to aid proteome annotation.
• Functional inference: Supports investigation of functional roles of parasite mitochondrial proteins for studies of parasite biology.
• Therapeutic target prioritization: Helps prioritize mitochondrial proteins as candidate targets for antimalarial research.

Methodology:

Computational methods explicitly include SVM models trained on a 175-protein dataset with five-fold cross-validation, amino acid composition analysis, Split Amino Acid Composition (SAAC), incorporation of PSSM profiles, and a hybrid PSSM+SAAC model.