PathInHydro, a Set of Machine Learning Models to Identify Unbinding Pathways of Gas Molecules in [NiFe] Hydrogenases

Area of Science:

• Biochemistry and Biophysics
• Computational Biology
• Enzyme Engineering

Background:

• [NiFe] hydrogenases are crucial for hydrogen (H2) production, a key future fuel.
• Enzyme sensitivity to oxygen (O2) and carbon monoxide (CO) necessitates protein engineering for stability.
• Understanding ligand unbinding mechanisms is vital for designing improved hydrogenases.

Purpose of the Study:

• To develop an automated method for analyzing molecular dynamics (MD) simulations of gas molecule dissociation from [NiFe] hydrogenases.
• To construct PathInHydro, a supervised machine learning model for identifying unbinding pathways.
• To facilitate faster and more efficient examination of MD simulation data.

Main Methods:

• Supervised machine learning models were trained using unbinding trajectories of CO and H2 from *Desulfovibrio fructosovorans* [NiFe] hydrogenase.
• PathInHydro was developed to automatically assign unbinding pathways for gas molecules.
• The model's performance was validated on different gas molecules and [NiFe] hydrogenase variants.

Main Results:

• PathInHydro successfully identified unbinding pathways for various gas molecules and [NiFe] hydrogenases.
• The model demonstrated feasibility for analyzing diverse gas-enzyme interactions, including mutated enzymes.
• Automated pathway assignment significantly reduces manual analysis time.

Conclusions:

• PathInHydro offers a robust and efficient solution for analyzing ligand unbinding in [NiFe] hydrogenase MD simulations.
• The tool accelerates research by automating complex trajectory analysis.
• This work supports the development of more stable and efficient hydrogenase enzymes for biotechnological applications.