Codon optimization with deep learning to enhance protein expression.

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• Heterologous expression is crucial for recombinant protein production in genetic synthesis.
• Current codon optimization methods rely on biological indexes, limiting their effectiveness.
• Developing advanced codon optimization strategies is essential for efficient protein production.

Purpose of the Study:

• To propose a novel codon optimization method utilizing deep learning.
• To enhance recombinant protein production through improved codon optimization.
• To validate the efficacy of the proposed deep learning method against existing approaches.

Main Methods:

• Introduced the concept of 'codon boxes' to recode DNA sequences.
• Framed codon optimization as a sequence annotation problem for amino acids.
• Trained Bidirectional Long-Short-Term Memory Conditional Random Field models for Escherichia Coli.
• Compared the deep learning method with the codon adaptation index and commercial optimization tools.

Main Results:

• The deep learning method demonstrated efficient enhancement of protein expression.
• Experimental results showed competitive performance against Genewiz and ThermoFisher optimizations.
• Protein expression experiments for Plasmodium falciparum candidate vaccine and polymerase acidic protein validated the method's effectiveness.

Conclusions:

• The proposed deep learning-based codon optimization method is efficient and competitive.
• This novel approach offers a promising alternative for improving recombinant protein production.
• Deep learning effectively captures DNA distribution characteristics for optimization.