Fitting quantum machine learning potentials to experimental free energy data: predicting tautomer ratios in solution.

Area of Science:

• Computational chemistry
• Drug discovery
• Quantum machine learning

Background:

• Tautomer ratios are critical for computer-aided drug discovery, as many approved drugs exist as multiple tautomeric species.
• Current quantum chemical methods for calculating aqueous tautomer ratios are computationally expensive and often inaccurate.
• These inaccuracies stem from the breakdown of rigid-rotor harmonic oscillator (RRHO) approximations.

Purpose of the Study:

• To investigate the limitations of current quantum chemical methods in calculating tautomer ratios.
• To introduce a novel quantum machine learning (QML) approach for accurate tautomer ratio computation.
• To demonstrate the ability of QML to overcome RRHO approximation limitations.

Main Methods:

• Utilized quantum machine learning (QML) for high-accuracy potential energy calculations at reduced cost.
• Employed rigorous relative alchemical free energy calculations.
• Developed tunable QML models trained on free energy data to correct underlying potential energy surfaces.

Main Results:

• Demonstrated that QML methods can compute tautomer ratios in vacuum, avoiding RRHO approximation limitations.
• Showcased the ability to generalize tautomer free energy predictions across diverse molecular structures.
• Achieved quantum chemical accuracy in energy calculations at a fraction of the computational cost.

Conclusions:

• Quantum machine learning provides a more accurate and efficient approach to calculating tautomer ratios for drug discovery.
• QML methods overcome the limitations of traditional RRHO approximations in complex molecular systems.
• The developed QML framework enables accurate prediction and generalization of tautomer free energies, advancing computer-aided drug discovery.