premier

premier predicts mannose-interacting residues in proteins to support analysis of mannose binding protein (MBP) recognition of pathogens.

Key Features:

• Machine learning: Support Vector Machine (SVM) models were used to predict mannose-interacting residues.
• Training set: Models were trained on 120 MBP chains with no more than 25% sequence similarity between any two chains.
• Dataset composition: A main dataset comprised 1,029 mannose interacting and 1,029 non-interacting residues, and a realistic dataset comprised 1,029 interacting and 10,320 non-interacting residues.
• Feature encodings: Predictions used binary patterns, position-specific scoring matrix (PSSM) profile patterns, and composition profiles as input encodings.
• Initial performance: Binary and PSSM profile patterns yielded an initial maximum Matthews Correlation Coefficient (MCC) of approximately 0.32.
• Refined performance: Composition profile–based models achieved an MCC of approximately 0.74 and accuracy of 86.64% on the main dataset.
• Performance on realistic dataset: On the realistic dataset the method achieved an MCC of 0.62 and accuracy of 93.08%.
• Compositional analysis: Analysis highlights preferred residue types for mannose interaction and identifies patterns in residues surrounding mannose interacting residues (MIRs).
• Generality: The compositional strategy suggests potential applicability to prediction of other types of interacting residues.

Scientific Applications:

• Protein annotation: Identification of MIRs to aid annotation of protein function related to mannose binding.
• MBP–pathogen interaction analysis: Elucidation of the structural basis of mannose binding protein recognition of pathogen surfaces.
• Immunology research: Investigation of mannose’s role in innate immune recognition and defense mechanisms.
• Method transfer: Application of the compositional prediction strategy to study other residue–ligand or residue–molecule interactions.

Methodology:

SVM models were trained on curated MBP chains (≤25% sequence similarity) using binary patterns, PSSM profile patterns, and composition profiles, with compositional analysis of residues surrounding MIRs.