MAHLER: Integrating Metadynamics and Inverse Folding to Predict Antibody-Antigen Kinetics

Area of Science:

• Biophysics
• Computational Biology
• Immunology

Background:

• Antibody binding kinetics significantly impact in vivo efficacy and pharmacokinetics.
• Current computational design methods primarily focus on equilibrium binding affinity, neglecting crucial kinetic parameters.

Purpose of the Study:

• To introduce MAHLER (Metadynamics-Anchored Hybrid Learning for Engineering off-Rates), a novel computational method for predicting antibody-antigen residence times.
• To provide a scalable and accurate tool for assessing antibody dissociation kinetics.

Main Methods:

• Developed a machine learning/physics hybrid approach integrating inverse-folding models with molecular dynamics simulations.
• MAHLER utilizes these simulations to predict relative antibody-antigen dissociation kinetics.

Main Results:

• MAHLER achieves screening-grade accuracy in calculating relative antibody-antigen dissociation kinetics for point mutants.
• Predictions are highly efficient, taking only 4 minutes per prediction on a single NVIDIA A100 GPU.
• Demonstrates a significant speed improvement over traditional enhanced molecular dynamics simulations.

Conclusions:

• MAHLER offers a practical, kinetics-aware complement to existing computational antibody design strategies.
• The method enables large-scale prediction of residence times, crucial for optimizing antibody function.
• Facilitates the development of antibodies with improved pharmacokinetic and in vivo efficacy profiles.