Predicting plant Rubisco kinetics from RbcL sequence data using machine learning.

Area of Science:

• Biochemistry
• Molecular Biology
• Computational Biology

Background:

• Ribulose-1,5-bisphosphate carboxylase/oxygenase (Rubisco) catalyzes CO2 fixation in photosynthesis, often limiting crop yield.
• Identifying superior Rubisco variants through natural diversity screening is crucial for crop engineering.

Purpose of the Study:

• To develop and apply machine learning models for predicting Rubisco enzyme kinetics from protein sequence data.
• To demonstrate the utility of Gaussian processes (GPs) for high-throughput screening of Rubisco variants.

Main Methods:

• Utilized Gaussian processes (GPs), a Bayesian modeling technique.
• Integrated GPs with protein encoding schemes to analyze Rubisco large subunit (RbcL) sequences.
• Trained models on existing experimental Rubisco kinetic data and applied them to over 9000 RbcL sequences.

Main Results:

• Successfully predicted Rubisco kinetic parameters from RbcL sequence data.
• Predicted kinetic values aligned with established trends, such as higher carboxylation turnover rates (Kcat) in C4/CAM species compared to C3 species.
• Demonstrated the feasibility of using machine learning for Rubisco kinetic prediction.

Conclusions:

• Machine learning, specifically Gaussian processes, offers a powerful tool for screening and predicting Rubisco kinetics.
• This computational approach can accelerate the discovery of improved Rubisco enzymes for enhanced photosynthesis and crop productivity.
• The methodology shows potential for application to other enzyme families beyond Rubisco.