GENErator

Torch implementation of a diffusion model designed for DNA sequence reconstruction and score prediction tasks, continuous, e.g. given a ChIP score, create a likely DNA sequence that might get it.

N.B. GRAHAM ON COMPUTE CANADA IS DOWN UNTIL JAN 7 -- THIS CODE IS NOT UP TO DATE AND MISSES A LOT OF STUFF THAT I CAN NOT ACCESS.

Features

• handle DNA sequences with variable lengths using padding and masking (up to a max)
• combined sequence reconstruction and score prediction objectives, helps to learn better

---

Installation

1. Clone the repository:

   git clone https://github.com/yourusername/dna-diffusion-model.git
   cd dna-diffusion-model
   pip install -e .

2. Data should look like. 1.23 ATCGTAA 0.89 GCTAGCTGCTA

3. Modify main to look like.

   data_generator = BEDFileDataGenerator(
    filepath='/path/to/your/dataset.bed', 
    num_sequences=15000, 
    maxlen=512
   )

4. Run like:

python main.py