akita-torch

PyTorch implementation of the Akita model from basenji, which is the code base for Fudenberg, Geoff, David R. Kelley, and Katherine S. Pollard. "Predicting 3D genome folding from DNA sequence with Akita." Nature methods 17.11 (2020): 1111-1117.

Installation

akita-torch can be installed via pip by

pip install akita-torch

Usage

from akita_torch.model import AkitaConfig, Akita

config = AkitaConfig()
model = Akita(config)

output = model(sample_1m_seq_1hot)  # output: (1, 99681, config.output_head_num)

Model Configuration

The original akita model configuration is implemented as the default values in AkitaConfig.

@dataclass
class AkitaConfig:
    output_head_num: int = 5
    target_crop: int = 32
    diagonal_offset: int = 2
    augment_rc: bool = True
    augment_shift: int = 11
    activation: str = "relu"
    norm_type: str = "batch"
    bn_momentum: float = 0.9265

Pretrained Akita Model

coming soon...