Prediction of membrane transport proteins and their substrate specificities using primary sequence information

Area of Science:

• Bioinformatics
• Computational Biology
• Molecular Biology

Background:

• Membrane transport proteins (transporters) are essential for cellular functions, moving substrates across membranes.
• Experimental characterization of transporters and their substrates is costly and time-consuming.
• Accurate prediction of transporter substrate specificity is an urgent bioinformatics task.

Purpose of the Study:

• To develop robust bioinformatics-based methods for predicting membrane transport protein substrate specificities.
• To differentiate transporters from non-transporter proteins.
• To provide a publicly accessible web server for transporter substrate specificity prediction.

Main Methods:

• Support Vector Machine (SVM)-based computational models were developed.
• Models integrated diverse protein sequence features: amino acid composition, dipeptide composition, physico-chemical composition, biochemical composition, and Position-Specific Scoring Matrices (PSSM).
• A hybrid model combining biochemical composition and PSSM was evaluated.

Main Results:

• The hybrid biochemical composition and PSSM model achieved 76.69% average prediction accuracy (AUC 0.833) on the main dataset.
• This model demonstrated strong performance on an independent dataset with 78.88% average accuracy.
• Evolutionary information (PSSM) and AAIndex were identified as key features for prediction.

Conclusions:

• Evolutionary information and biochemical features are critical for accurate transporter substrate specificity prediction.
• Similarity-based methods (BLAST, PSI-BLAST, HMMs) showed lower predictive accuracy.
• The TrSSP web server, implementing these SVM models, is now available for public use.