Integrating Machine Learning and Quantum Circuits for Proton Affinity Predictions.

Area of Science:

• Computational Chemistry
• Machine Learning
• Quantum Computing

Background:

• Interpreting gas-phase ion mobility-mass spectrometry (IM-MS) data for structure prediction requires identifying the most favorable protonated site.
• Proton affinity (PA) measurements determine the site of protonation, but current methods (mass spectrometry, ab initio computation) are resource-intensive and time-consuming.
• Efficient PA estimation is needed for rapid identification of proton binding sites in complex organic molecules.

Purpose of the Study:

• To develop a fast and accurate method for predicting proton affinity (PA).
• To explore the potential of hybrid quantum-classical machine learning models for PA prediction.

Main Methods:

• Developed a machine learning (ML) model using 186 molecular descriptors for PA prediction.
• Designed quantum circuits as feature encoders for a classical neural network, creating a hybrid quantum-classical model.
• Compared the performance of the ML model and the hybrid model against traditional methods using reduced feature sets.

Main Results:

• The ML model achieved high predictive performance with an R² of 0.96 and a MAE of 2.47 kcal/mol.
• Quantum-encoded features showed stronger positive correlation with target PA values than original features.
• The hybrid quantum-classical model outperformed its classical counterpart and demonstrated performance comparable to traditional ML models.

Conclusions:

• The developed ML model provides a fast and accurate method for PA prediction.
• Hybrid quantum-classical models show significant potential for enhancing the accuracy and efficiency of PA predictions.
• This work highlights the promise of quantum machine learning in computational chemistry applications.