Codon Optimization Using a Recurrent Neural Network.

Area of Science:

• Biotechnology
• Molecular Biology
• Bioinformatics

Background:

• Codon optimization increases protein expression efficiency for biologic pharmaceuticals.
• Traditional methods like codon usage bias and GC content have limitations.
• Machine learning may uncover novel patterns for improved optimization.

Purpose of the Study:

• To explore undirected codon optimization using machine learning.
• To develop and test a recurrent neural network (RNN) model for DNA sequence optimization.
• To compare RNN-based optimization against conventional algorithms.

Main Methods:

• Trained a recurrent neural network (RNN) model on Chinese hamster DNA sequences.
• Generated optimized DNA sequences for programmed death-ligand 1 and a monoclonal antibody.
• Transfected RNN-optimized and conventionally optimized sequences into Chinese hamster ovary cells.

Main Results:

• RNN-optimized DNA sequences resulted in protein expression levels equal to or higher than conventionally optimized sequences.
• Demonstrated the efficacy of machine learning in codon optimization.
• Validated the RNN model's performance in a cellular context.

Conclusions:

• Machine learning, specifically RNNs, offers a powerful approach to codon optimization.
• This method can enhance protein expression efficiency for biopharmaceutical manufacturing.
• Undirected optimization holds potential for discovering new patterns in DNA sequences.