Codon optimization • cubar

Codon optimization modifies a gene’s coding sequence to enhance protein production without changing the encoded amino acid sequence. The codon_optimize function in the cubar package provides three strategies for optimizing coding sequences based on the codon usage of the target organism. The first two strategies replace rare codons with more frequently used ones, while the third strategy employs the third-party deep learning model CodonTransformer (Fallahpour et al., 2025) to optimize codon usage. Additionally, cubar can integrate the state-of-the-art deep learning model SpliceAI (Jaganathan et al., 2019) to prevent the unintended introduction of cryptic splice sites in optimized sequences. As both SpliceAI and CodonTransformer are Python-based, users must manually install these packages. Here we demonstrates how to install them using conda (or mamba) in a new environment for use in cubar.

# create a new environment named "cubar_env" with both python and r installed
conda create -n cubar_env python=3.12 r-base blas=*=netlib r-reticulate
# activate the environment we just created
conda activate cubar_env
# install CodonTransformer and SpliceAI
pip install CodonTransformer tensorflow spliceai

“naive” method

The default “naive” method simply replaces each codon to the most preferred one in the same family or subfamily.

library(cubar)

seq <- 'ATGCTACGA'
cf_all <- count_codons(yeast_cds)
#> Loading required namespace: Biostrings
optimal_codons <- est_optimal_codons(cf_all)
seq_opt <- codon_optimize(seq, optimal_codons)
print(seq_opt)
#> 9-letter DNAString object
#> seq: ATGCTACGT

“IDT” method

The “IDT” option implements the method used by the codon optimization tool of Integrated DNA Technologies. Briefly, this method randomly selects synonymous codons from the same family or subfamily based on their relative frequency, but excluding rare codons used below 10% in the target organism.

seq_opt <- codon_optimize(seq, cf = cf_all, method = "IDT")
print(seq_opt)
#> 9-letter DNAString object
#> seq: ATGCTGCGA

“CodonTransformer” method

The “CodonTransformer” method optimizes codon usage with the third-party software CodonTransformer directly using a wrapper in R. CodonTransformer is a deep learning model that can generate coding sequences that show similar codon usage and distribution to host genes with reduced negative cis elements in a wide range of organisms across the tree of life. Please refer to the original study for more details.

seq_opt <- codon_optimize(seq, method = "CodonTransformer", organism = "Saccharomyces cerevisiae")
print(seq_opt)

cubar can generate several optimized sequences at the same time using the argument num_sequences with the method “IDT” and “CodonTransformer”. When num_sequences is greater than 1, identical duplicate sequences will be retained as a single copy, potentially resulting in a final sequence count less than the specified value.

seqs_opt <- codon_optimize(seq, cf = cf_all, method = "IDT", num_sequences = 10)
print(seqs_opt)
seqs_opt <- codon_optimize(seq, method = "CodonTransformer", organism = "Saccharomyces cerevisiae",
num_sequences = 10, deterministic =FALSE, temperature = 0.4)
print(seqs_opt)

Splice site detection

In addition, cubar integrated the deep learning tool SpliceAI to identify potential splice sites with the argument spliceai. When the probabilities of non-splice site for each base are greater than 0.5, it is considered that there are no potential splice junction sites, and the Possible_splice_junction in the output is marked as FALSE, otherwise it is marked as TRUE.

seqs_opt <- codon_optimize(seq, cf = cf_all, method = "IDT", num_sequences = 10, spliceai = TRUE)
print(seqs_opt)
seq_opt <- codon_optimize(seq, method = "CodonTransformer", organism = "Saccharomyces cerevisiae", spliceai = TRUE)
print(seq_opt)

References

Fallahpour A, Gureghian V, Filion GJ, Lindner AB, Pandi A. CodonTransformer: a multispecies codon optimizer using context-aware neural networks. Nat Commun. 2025 Apr 3;16(1):3205. doi: 10.1038/s41467-025-58588-7. PMID: 40180930; PMCID: PMC11968976.
Jaganathan K, Kyriazopoulou Panagiotopoulou S, McRae JF, Darbandi SF, Knowles D, Li YI, Kosmicki JA, Arbelaez J, Cui W, Schwartz GB, Chow ED, Kanterakis E, Gao H, Kia A, Batzoglou S, Sanders SJ, Farh KK. Predicting Splicing from Primary Sequence with Deep Learning. Cell. 2019 Jan 24;176(3):535-548.e24. doi: 10.1016/j.cell.2018.12.015. Epub 2019 Jan 17. PMID: 30661751.
Method used by the IDT codon optimization tool: https://sg.idtdna.com/pages/education/decoded/article/idt-codon-optimization-tool-makes-synthetic-gene-design-easy