Improving Protein Expression Prediction Using Extra Features and Ensemble Averaging.

Area of Science:

• Computational Biology
• Bioinformatics
• Synthetic Biology

Background:

• Predicting protein expression levels from codon sequences is crucial for optimizing protein production.
• Existing machine learning models utilize features like codon bias and minimum free energy.

Purpose of the Study:

• To enhance machine learning models for predicting protein expression levels.
• To investigate the impact of additional input features and ensemble methods on model performance.

Main Methods:

• Support Vector Regression (SVR) and Partial Least Squares (PLS) were employed for model development.
• New features, including codon identification number and codon count, were incorporated.
• Ensemble averaging techniques were applied to existing models.

Main Results:

• Support Vector Regression (SVR) models demonstrated superior predictive performance compared to Partial Least Squares (PLS).
• The inclusion of codon identification number and codon count significantly improved model predictive ability.
• Ensemble averaging further enhanced the accuracy of both SVR and PLS models.

Conclusions:

• Machine learning models can be effectively improved for protein expression prediction by incorporating additional codon-derived features and ensemble techniques.
• Further research into diverse ensembles and features is warranted to achieve near-perfect prediction accuracy.
• These findings offer valuable insights for optimizing codon usage and boosting protein expression in synthetic biology.