Identification of ligand binding pockets on nuclear receptors by machine learning methods

Area of Science:

• Biochemistry
• Molecular Biology
• Pharmacology

Background:

• Nuclear receptors are crucial transcription factors regulating gene expression.
• They are targeted by 13% of drugs, highlighting the importance of understanding their ligand binding pockets.
• Identifying these pockets is key for designing effective therapeutics.

Purpose of the Study:

• To develop a computational method for identifying nuclear receptor ligand binding pockets.
• To classify nuclear receptors into different classes for class-specific drug design.
• To identify key features of nuclear receptor binding pockets for drug screening.

Main Methods:

• Support Vector Machine (SVM) classifier was employed.
• Datasets included known nuclear receptor-ligand complex structures and control non-binding pockets.
• A multiclass SVM model was used for nuclear receptor classification.

Main Results:

• SVM model achieved 96% 10-fold cross-validation accuracy for identifying ligand binding pockets.
• Multiclass SVM model yielded 92% average 10-fold cross-validation accuracy for receptor classification.
• Top features indicated hydrophobic pockets with conserved Leucine and phenylalanine residues.

Conclusions:

• SVM is a highly accurate tool for identifying and classifying nuclear receptor binding pockets.
• Key features of these pockets, like hydrophobicity and specific residues, are identified.
• This work facilitates the screening of drug molecules targeting nuclear receptors.