DELIA

DELIA predicts protein-ligand binding sites using a hybrid deep neural network that integrates 1D sequence-based features and 2D structure-based amino acid distance matrices to distinguish binding from non-binding residues.

Key Features:

• Hybrid deep neural network: Integrates 1D sequence-based features with 2D structure-based amino acid distance matrices to combine linear and spatial protein representations.
• Class imbalance handling: Employs oversampling within mini-batches, random undersampling, and stacking ensemble techniques to mitigate severe imbalance between binding and nonbinding residues.
• Comparison to alignment methods: Addresses limitations of structure-alignment-based methods that depend on annotated homogeneous protein structures.
• Benchmark validation: Demonstrated superior performance on five benchmark datasets for predicting protein-ligand binding residues.

Scientific Applications:

• Protein function and interaction analysis: Identifies ligand-binding residues to inform understanding of protein functions and interaction mechanisms.
• Drug discovery: Provides residue-level predictions that can guide structure-based drug discovery and ligand design.
• High-throughput prediction: Enables large-scale prediction of protein-ligand binding residues across diverse protein datasets.

Methodology:

DELIA uses a hybrid deep neural network combining 1D sequence-based features and 2D structure-based amino acid distance matrices, and addresses class imbalance via oversampling within mini-batches, random undersampling, and stacking ensemble techniques, with validation on five benchmark datasets.