Hyperbolic Genome Embeddings

Supplementary code for the paper Hyperbolic Genome Embeddings. The hyperbolic models used in this paper rely on the library of network components for HNNs in the Lorentz model detailed in the HyperbolicCV repository.

Installation

Requirements

• Python>=3.8

conda create -n HGE python=3.8

• The repo is developed with Pytorch 1.13, using cuda 11.7

conda install pytorch==1.13.0 torchvision==0.14.0 torchaudio==0.13.0 pytorch-cuda=11.7 -c pytorch -c nvidia

• install requirements:

pip install -r requirements.txt

• additionally, we use the embedders repo to sample points in our exploration of hyperbolicity.

Training

• For classification, choose a config file and adapt the following command.

  python train.py -c configs/HCNN_SingleK_TEB.txt

  You can also add additional arguments without changing the config file. For example:

  python train.py -c configs/HCNN_SingleK_TEB.txt\
     --output_dir classification/output --device cuda:1 --dataset dna_cmc --length 200

Transposable Elements Benchmark (TEB)

Download the TEB datasets from here. Please see the HEB paper for details on dataset construction. Summary statistics for TEB: