DeepFrag

DeepFrag predicts molecular fragments to extend ligands within receptor/ligand complexes to improve binding affinity for lead optimization in computer-aided drug discovery.

Key Features:

• Fragment Prediction: Uses a deep convolutional neural network to analyze structural data of receptor/ligand complexes and suggest molecular fragments for ligand extension.
• Benchmark Performance: In independent tests with intentionally deleted fragments, it identified the correct fragment from a pool of over 6,500 options approximately 58% of the time.
• Chemical Similarity of Predictions: When the exact known fragment was not selected, the top predicted fragments were often chemically similar to the original, providing viable alternative substitutions.

Scientific Applications:

• Lead Optimization: Proposes fragment-based modifications to refine known ligands and enhance ligand–receptor binding affinity.
• Binding-Affinity Prediction: Supports prediction of improved ligand–receptor interactions by recommending fragment extensions expected to increase affinity.
• Virtual Screening Integration: Can be integrated into virtual screening workflows to suggest fragment additions for candidate ligands.

Methodology:

A deep convolutional neural network trained on structural datasets of receptor/ligand complexes predicts and ranks complementary fragment extensions.