Calculating Protein-Ligand Residence Times through State Predictive Information Bottleneck Based Enhanced Sampling.

Area of Science:

• Biochemistry and computational drug discovery.
• Molecular dynamics simulations and machine learning applications.

Background:

• Drug residence time is critical for efficacy, yet difficult to study at the atomic level using traditional molecular dynamics (MD) simulations due to extreme time scales.
• Predicting protein-ligand residence times accurately remains a significant challenge in computational biochemistry.

Purpose of the Study:

• To develop and validate a semi-automated computational protocol for calculating ligand residence times over 12 orders of magnitude.
• To integrate deep learning with enhanced sampling methods to overcome limitations in simulating long time scales.

Main Methods:

• Developed a semi-automated protocol integrating a deep learning method, the state predictive information bottleneck (SPIB), to approximate reaction coordinates.
• Utilized the learned reaction coordinate to guide the metadynamics enhanced sampling method for efficient simulation.
• Applied the protocol to six protein-ligand complexes, including Imatinib (Gleevec) with Abl kinase and its mutants.

Main Results:

• The protocol successfully recovered quantitatively accurate ligand residence times across a wide range of time scales.
• Demonstrated the method's efficacy on benchmark systems with known residence times, including drug-resistant mutants.
• Validated the ability to predict residence times for the anticancer drug Imatinib.

Conclusions:

• The developed protocol offers a powerful and accurate approach for calculating ligand residence times.
• This method has the potential to provide deeper insights into drug development and ligand recognition mechanisms.
• Advances in rare event sampling and deep learning enable accurate prediction of crucial drug-target interaction dynamics.