CODONIFY: A Recurrent-Neural-Network-based Codon Optimization Tool to Improve Protein Expression Towards Efficient Vaccine Manufacturing

Overview

Designing synthetic genes for heterologous expression is a keystone of biotechnology. In protein sequences—as there are 61 sense codons but only 20 standard amino acids—most amino acids are encoded by more than one codon. Although such synonymous codons do not alter the encoded amino acid sequence, they are not redundant. Industry-standard codon optimization techniques based on biological indexes replace synonymous codons with the most abundant codon found in the host organism’s genome. However, this technique may result in an imbalanced tRNA pool, metabolic stress, and translational error which lead to greater cell toxicity and reduced protein expression. In this research, recurrent neural networks are used to accurately capture sequential and contextual patterns. By predicting synonymous codons based on the sequential information of the host organism, protein expression can be increased while preventing translational error and plasmid toxicity. The model uses a bidirectional long short-term memory-based architecture, allowing for the host genome to be taken into context. The Codon Adaptation Index (CAI) was used to measure synonymous codon usage. When tested on eGFP and FALVAC-1, the model yielded a 0% mutation rate and improved CAI from 0.72 and 0.67 to 0.91 and 0.91 respectively. On a broad test dataset of 8,000 sequences, Codonify optimized CAI by 22% which correlates with an average 236% increase in expression. This research provides evidence that sequential context may yield codon selection that is more similar to the host genome, therefore increasing protein expression and the production of recombinant vaccines.

Awards (1)

• Third Award of $1,000 $1,000

Competition history

• ISEF 2021 Biomedical Engineering · Entry ENBM043

Resources

• ISEF abstract page ↗

Related projects

Closest projects by meaning, across every fair and year in the corpus.