MemType-2L

MemType-2L classifies protein sequences by detecting membrane proteins and assigning membrane proteins to one of eight types (type I, type II, type III, type IV, multipass, lipid-chain-anchored, GPI-anchored, and peripheral) to support membrane-protein functional characterization.

Key Features:

• Two-Layer Predictive Framework: A two-step classifier first discriminates membrane versus non-membrane proteins and then assigns membrane proteins to one of eight specific types: type I, type II, type III, type IV, multipass, lipid-chain-anchored, GPI-anchored, and peripheral.
• Pseudo Position-Specific Score Matrix (Pse-PSSM): Represents protein samples using Pse-PSSM vectors to incorporate evolutionary conservation into the feature set.
• Ensemble OET-KNN Classifier: Uses an ensemble composed of multiple optimized Evidence-Theoretic K-Nearest Neighbor (OET-KNN) classifiers for robust prediction.
• Performance Evaluation: Prediction performance was evaluated using jackknife tests and independent dataset tests.

Scientific Applications:

• Proteome Annotation: Annotating proteomes by identifying membrane proteins and their specific membrane-type classes for downstream analyses.
• Functional Inference: Inferring potential biological roles and interaction mechanisms of membrane proteins from predicted types.
• High-Throughput Classification: Enabling large-scale classification of uncharacterized protein sequences in post-genomic datasets.

Methodology:

Protein sequences are represented by Pse-PSSM vectors; a two-step prediction first filters membrane versus non-membrane proteins and then classifies membrane proteins into eight types using an ensemble of optimized Evidence-Theoretic K-Nearest Neighbor (OET-KNN) classifiers, with performance assessed by jackknife and independent dataset tests.