MONN

MONN predicts non-covalent interactions and binding affinities between compounds and proteins to improve interpretability of interaction sites and support molecular-level analysis for drug discovery.

Key Features:

• Multi-Objective Prediction: MONN simultaneously predicts non-covalent interactions and binding affinities and leverages their joint prediction to improve accuracy.
• Convolutional Neural Networks (CNNs): It employs convolutional neural networks (CNNs) on molecular graphs representing compounds and on primary sequences of proteins to extract intrinsic features.
• Interpretability: MONN addresses interpretability limitations of neural attention-based models by systematically evaluating local interaction sites without requiring additional supervision.
• Benchmark Dataset: Development and evaluation used a benchmark dataset containing over 10,000 compound–protein pairs with documented non-covalent interactions.
• Performance Evaluation: MONN outperforms state-of-the-art methods in binding affinity prediction and predicts non-covalent interactions on both the benchmark dataset and an independent dataset derived from the Protein Data Bank (PDB).
• Predictive Power and Insights: Predicted pairwise interactions exhibit patterns consistent with chemical properties, providing molecular-level insights relevant to drug discovery.

Scientific Applications:

• Compound–Protein Interaction Analysis: MONN provides accurate predictions of non-covalent interactions and binding affinities to elucidate molecular mechanisms of compound–protein interactions.
• Therapeutic Target Identification and Lead Optimization: MONN’s predictions assist in identifying therapeutic targets and optimizing drug candidates.

Methodology:

Convolutional neural networks process molecular graphs for compounds and primary protein sequences to jointly predict non-covalent interactions and binding affinities.