tmbj-aidd/aptagpt-bcma

AptaGPT

AptaGPT is a generative pre-trained language model for aptamer design. The model focuses on the generation of a new sequence space of aptamers, trained and fine-tuned using the third and sixth round of SELEX data for B cell maturation antigen (BCMA).

Dataset

AptaGPT was pre-trained using a large dataset consisting of 108,229,900 sequences from the third round of the SELEX process targeting BCMA. This extensive dataset provided a robust foundation for learning generalized patterns in aptamer sequences. For fine-tuning, the model utilized 9,350 sequences from the sixth round of SELEX. All aptamer sequences used for both pre-training and fine-tuning are 35 nucleotides in length.

Requirements

Before running the AptaGPT model, the following Python dependencies need to be installed:

pip install transformers sentencepiece

Usage Examples

To load the model form hugging face:

from transformers import pipeline
aptagpt = pipeline('text-generation', model="tmbj-aidd/aptagpt-bcma")

To generate aptamer sequences:

sequences = aptagpt("<|endoftext|>",
                max_length=15,
                do_sample=True,
                top_k=700,
                repetition_penalty=1.2,
                num_return_sequences=10,
                )
print(sequences)