K_DEEP: Protein-Ligand Absolute Binding Affinity Prediction via 3D-Convolutional Neural Networks

Area of Science:

• Computational chemistry
• Drug discovery
• Machine learning

Background:

• Accurate prediction of protein-ligand binding affinities is crucial for accelerating drug discovery.
• Existing methods often lack the speed and accuracy required for large-scale virtual screening and lead optimization.

Purpose of the Study:

• To introduce K_DEEP, a novel machine learning approach for predicting protein-ligand binding affinities.
• To evaluate the performance of K_DEEP against other computational methods using diverse datasets.

Main Methods:

• Utilized state-of-the-art 3D-convolutional neural networks for binding affinity prediction.
• Compared K_DEEP with existing machine learning and scoring functions on standard datasets like PDBbind (v.2016).

Main Results:

• K_DEEP achieved state-of-the-art performance on the PDBbind core test set, with a Pearson's correlation coefficient of 0.82 and RMSE of 1.27 pK units.
• While highly accurate, the model's performance showed sensitivity to the specific protein target.
• Each prediction by K_DEEP takes only a fraction of a second.

Conclusions:

• K_DEEP offers a fast and accurate machine learning solution for predicting protein-ligand binding affinities.
• Its speed, performance, and ease of use make it a valuable tool for computational chemistry pipelines.
• K_DEEP is accessible via PlayMolecule.org for practical application in drug discovery research.