druggable

druggable predicts druggable proteins using machine learning and functional enrichment analysis to identify proteins capable of binding antibodies or small molecules based on amino acid composition descriptors.

Key Features:

• Machine Learning Approach: Uses linear and non-linear classifiers to analyze protein sequences.
• Amino Acid Composition Descriptors: Represents proteins with 200 tri-amino acid composition descriptors.
• Feature Selection and Classifier Optimization: Performs rigorous feature selection and identifies a Support Vector Machine (SVM) classifier as the most effective model.
• Predictive Performance: Reports AUROC 0.975 ± 0.003 and accuracy 0.929 ± 0.006 using three-fold cross-validation.
• Functional Enrichment Analysis: Integrates functional enrichment analysis to prioritize proteins that bind antibodies or small molecules with suitable chemical properties and affinity.

Scientific Applications:

• Breast cancer protein set: Identifies top predicted druggable proteins including CDK4, AP1S1, and POLE.
• Cancer-driving protein set: Highlights predicted druggable proteins such as TLL2, FAM47C, and MACC1.
• RNA-binding protein set: Predicts druggable proteins including PLA2G1B, CPEB2, and NOL6.
• Therapeutic target prioritization: Facilitates prioritization of promising therapeutic targets to inform drug discovery efforts.

Methodology:

Builds a prediction model from protein sequences using 200 tri-amino acid composition descriptors, applies linear and non-linear machine learning classifiers with feature selection (selecting SVM as best), and evaluates performance by three-fold cross-validation reporting AUROC and accuracy; includes functional enrichment analysis.