DeepCodon: A deep learning codon-optimization model to enhance protein expression.

Area of Science:

• Synthetic Biology
• Genetic Engineering
• Computational Biology

Background:

• Codon optimization is crucial for heterologous gene expression, involving complex factors like host codon bias and mRNA structure.
• Existing methods often overlook functionally important rare codon clusters, limiting optimization effectiveness.
• Developing advanced tools is necessary to address the multiobjective nature of codon optimization.

Purpose of the Study:

• Introduce DeepCodon, a novel deep learning tool for codon optimization.
• Focus on preserving functionally important rare codon clusters often missed by conventional approaches.
• Enhance heterologous gene expression by improving sequence design.

Main Methods:

• Trained a protein-CDS translation model on 1.5 million Enterobacteriaceae sequences, fine-tuned with highly expressed genes.
• Integrated a conditional probability strategy to preserve conserved rare codons.
• Utilized Escherichia coli as the host for gene expression and model training.

Main Results:

• DeepCodon generates sequences that better align with host preferences and exhibit superior in silico metrics.
• The tool successfully preserves critical rare codons, a key differentiator from traditional methods.
• Experimental validation showed DeepCodon outperformed conventional approaches in 9 out of 20 cases (7 P450s, 13 G3PDHs).

Conclusions:

• DeepCodon offers a practical and effective solution for codon optimization challenges.
• The tool demonstrates significant potential for improving heterologous gene expression in synthetic biology.
• Preserving rare codon clusters is a viable strategy for enhancing protein production.