Machine Learning-Guided Optimization of SABRE Hyperpolarization for α-Ketoglutarate in Acetone-Water.

Area of Science:

• Nuclear Magnetic Resonance (NMR) spectroscopy
• Hyperpolarization techniques
• Metabolic imaging

Background:

• Signal Amplification by Reversible Exchange (SABRE) is an efficient hyperpolarization method.
• Recent advances like Ace-SABRE enable biocompatible solutions for metabolic monitoring.
• Expanding substrate scope for Ace-SABRE is crucial for broader applications.

Purpose of the Study:

• To apply machine learning (ML), specifically Bayesian optimization (BO), to accelerate SABRE optimization.
• To model complex SABRE dynamics and reduce experimental time.
• To gain chemical insights for predicting improved sample compositions.

Main Methods:

• Bayesian optimization (BO) applied to four key input parameters of SABRE.
• Development of an ML model to simulate SABRE dynamics.
• Experimental validation of ML-guided parameter optimization.

Main Results:

• Achieved a maximum observed free polarization of ~6.6% for 1-¹³C alpha-ketoglutarate (AKG).
• Significantly improved upon the original average free polarization of ~0.90%.
• ML model provided chemical insights leading to enhanced polarization.

Conclusions:

• ML-driven optimization is a powerful tool for accelerating SABRE development.
• This approach enables efficient exploration of SABRE parameter space.
• The developed model facilitates increased polarization for metabolic monitoring.