MEMSAT-SVM

MEMSAT-SVM predicts transmembrane protein topology and identifies pore-lining regions from sequence using support vector machines to characterize alpha-helical channels and transporters.

Key Features:

• SVM-based topology and pore-lining prediction: A support vector machine classifier predicts transmembrane topology and the likelihood that a helix lines a pore from sequence data alone.
• Pore-lining residue labeling: Automatically identifies pore-lining residues using a labeling method based on geometric criteria derived from crystal structures.
• Support vector regression for stoichiometry: Uses support vector regression to predict the number of subunits participating in the pore.
• Performance metrics: Achieves 72% accuracy for helix pore-lining prediction and 62% accuracy for subunit-number prediction under stringent cross-validation.
• Targeted protein class: Focused on alpha-helical transmembrane channel and transporter proteins.

Scientific Applications:

• Pore characterization: Characterizing pore-lining regions in transmembrane proteins to support structure–function analyses.
• Pore stoichiometry inference: Providing predictions of the number of subunits in pores to inform hypotheses about oligomeric state.
• Channel and transporter studies: Supporting investigation of alpha-helical channels and transporters that are under-represented in structural databases.
• Therapeutic research support: Informing studies that seek to target transmembrane pores for therapeutic intervention.

Methodology:

Labels pore-lining residues using geometric criteria derived from crystal structures, trains a support vector machine classifier on sequence information to predict helix pore-lining likelihood, and employs support vector regression to predict pore subunit number with performance assessed by stringent cross-validation.