Supervised fine-tune of PlasmidGPT on a curated corpus of ~15k engineered E. coli plasmids from PlasmidScope and Addgene (Cunningham et al., 2025). Used as a baseline for the GRPO-trained PlasmidGPT-GRPO.

Quick start

python

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("UCL-CSSB/PlasmidGPT-SFT")
tokenizer = AutoTokenizer.from_pretrained("UCL-CSSB/PlasmidGPT-SFT")

input_ids = tokenizer("ATG", return_tensors="pt").input_ids
outputs = model.generate(input_ids, max_new_tokens=512, do_sample=True, temperature=1.0)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Citation

bibtex

@article{cunningham2025plasmidsft,
  title   = {Generative design and construction of functional plasmids with a {DNA} language model},
  author  = {Cunningham, Angus G. and Dekker, Linda and Shcherbakova, Anastasiia and Barnes, Chris P.},
  journal = {bioRxiv},
  year    = {2025},
  doi     = {10.64898/2025.12.06.692736}
}