MPLs-Pred: Predicting Membrane Protein-Ligand Binding Sites Using Hybrid Sequence-Based Features and Ligand-Specific...

Area of Science:

• Biochemistry
• Computational Biology
• Pharmacology

Background:

• Membrane proteins (MPs) are crucial for biological transport and are key targets for drug development.
• Identifying membrane protein-ligand binding sites (MPLs) is vital for efficient drug discovery.

Purpose of the Study:

• To develop an accurate, sequence-based predictor for identifying membrane protein-ligand binding sites (MPLs).
• To enhance drug discovery by improving the prediction of interactions with membrane proteins.

Main Methods:

• MPLs-Pred utilizes a random forest classifier integrating evolutionary profiles, topology structure, physicochemical properties, and sequence segment descriptors.
• An under-sampling scheme addresses imbalanced sample classification.
• Ligand-specific models were incorporated for refined prediction.

Main Results:

• MPLs-Pred achieved an overall Matthews correlation coefficient (MCC) of 0.63.
• Specific MCC values were 0.604 for drugs, 0.7 for metal ions, and 0.692 for biomacromolecules.
• The method demonstrates appreciable performance in predicting MPLs.

Conclusions:

• MPLs-Pred offers a valuable computational tool for predicting membrane protein-ligand binding sites.
• The predictor can significantly contribute to accelerating drug discovery and development pipelines.
• The tool is freely accessible for research use.