GPCRpred

GPCRpred predicts the family and subfamily classification of G-protein coupled receptors (GPCRs) from protein sequences using support vector machines (SVMs) and dipeptide composition to enable sequence-based annotation and therapeutic target identification.

Key Features:

• Methodology: Uses an SVM-based approach that leverages dipeptide composition of protein sequences for GPCR versus non-GPCR discrimination and for family and subfamily classification.
• Multiclass Classification: Employs multiclass SVMs for recognition and classification across multiple GPCR classes and subfamilies.
• Training Dataset: Developed using a dataset sourced from http://www.soe.ucsc.edu/research/compbio/gpcr/.
• Validation: Evaluated using 5-fold cross-validation to assess robustness and reliability.
• Performance — GPCR Detection: Achieves an overall accuracy of 99.5% for distinguishing GPCRs from non-GPCRs.
• Performance — Five Major Classes: Classifies five major GPCR classes with a Matthew's correlation coefficient (MCC) of 0.81 and an accuracy of 97.5%.
• Performance — Rhodopsin-like Subfamilies: Identifies subfamilies within the rhodopsin-like family with an MCC of 0.97 and an accuracy of 97.3%.
• Performance — Independent Blind Test: Demonstrates accuracies of 91.3% at the family level and 96.4% at the subfamily level on a blind dataset of 650 GPCRs.
• Contribution to Classification: Proposed subfamily assignments for 42 sequences previously classified as unclassified Class A GPCRs.

Scientific Applications:

• Drug Design: Assists in identifying and prioritizing GPCR targets for therapeutic intervention by providing family and subfamily annotations.
• Protein Classification: Supports annotation of newly discovered protein sequences as GPCRs and assignment to specific families and subfamilies.
• Biological Research: Facilitates studies of GPCR-mediated cellular signaling by supplying sequence-based classification information.

Methodology:

Support vector machine (SVM)-based classification using dipeptide composition with multiclass SVMs, evaluated by 5-fold cross-validation.