A deep learning model trained on expressed transcripts across different tissue types reveals cell-type...

Area of Science:

• Biotechnology
• Computational Biology
• Molecular Biology

Background:

• Species-specific protein translation differences necessitate codon optimization for recombinant protein expression.
• Existing codon optimization tools can be ineffective, leading to reduced protein expression or misfolding.

Purpose of the Study:

• To develop a novel deep learning (DL) tool utilizing a recurrent neural network (RNN) for cell type-dependent codon bias prediction.
• To improve the efficiency and accuracy of codon optimization for enhanced protein expression.

Main Methods:

• Trained DL models using gene expression data from brain, liver, and muscle tissues for secretory genes.
• Developed RNN-based models to predict optimal codon usage specific to cell types.
• Evaluated codon-optimized sequences using reporter genes in vitro.

Main Results:

• Codon-optimized sequences generated by the DL tool showed significantly enhanced protein expression compared to original and conventionally optimized sequences.
• DL models trained on liver cell gene expression data yielded the highest in vitro expression, regardless of the tested cell type.
• The DL approach proved effective in enhancing protein translation, especially for secretory proteins.

Conclusions:

• Deep learning-based codon optimization offers a significant advancement over existing methods.
• This novel approach has broad implications for the production of protein-based pharmaceuticals, vaccines, and gene therapy products.
• Cell type-specific codon bias prediction using DL can overcome limitations of current optimization strategies.