calvinmoody1/codontransformer-adp1

CodonTransformer — Acinetobacter baylyi ADP1 Fine-Tuned Checkpoint

Fine-tuned checkpoint of CodonTransformer with an additional organism embedding for Acinetobacter baylyi ADP1 (organism ID 164), produced by ASU iGEM 2026.

What is this?

CodonTransformer is a BERT-based model that takes a protein amino acid sequence and predicts synonymous DNA codons optimized for a target organism's codon usage bias.

The published base model supports 164 organisms. This checkpoint extends it with a fine-tuned embedding for Acinetobacter baylyi ADP1, the chassis used in our iGEM biosensor project for detecting Mycobacterium tuberculosis DNA via a dCas9-based reporter.

Checkpoint file

File	Description
a_baylyi_adp1/a_baylyi_adp1.ckpt	PyTorch Lightning checkpoint, 12 training epochs on ADP1 codon usage data

Usage

With the ASU iGEM wiki

Set these variables on your Hugging Face Space:

CODONTRANSFORMER_MODEL_REPO_ID=calvinmoody1/codontransformer-adp1
CODONTRANSFORMER_MODEL_FILENAME=a_baylyi_adp1/a_baylyi_adp1.ckpt

Locally

from CodonTransformer.CodonPrediction import load_model, load_tokenizer, predict_dna_sequence
from CodonTransformer.CodonData import get_amino_acid_sequence
from huggingface_hub import hf_hub_download
import torch

ckpt_path = hf_hub_download(
    repo_id="calvinmoody1/codontransformer-adp1",
    filename="a_baylyi_adp1/a_baylyi_adp1.ckpt",
)

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
tokenizer = load_tokenizer()
model = load_model(model_path=ckpt_path, device=device)

protein = "MKTIIALSYIFCLVFA*"  # must end with stop symbol
result = predict_dna_sequence(
    protein=protein,
    organism="Acinetobacter baylyi ADP1",
    device=device,
    tokenizer=tokenizer,
    model=model,
    deterministic=True,
    match_protein=True,
)
print(result.predicted_dna)

Training

Fine-tuned from the CodonTransformer base checkpoint using ADP1 codon usage data compiled by the ASU iGEM 2026 team. Training was performed for 12 epochs.

Citation

If you use this checkpoint, please cite the original CodonTransformer paper:

@article{Fallahpour2024,
  title={CodonTransformer: a multispecies codon optimizer using context-aware neural networks},
  author={Fallahpour, Adibvafa and others},
  journal={Nature Machine Intelligence},
  year={2024},
  doi={10.1038/s42256-024-00961-2}
}

License

MIT — same as the upstream CodonTransformer project.