How to train Region2Vec interval embeddings

Region2Vec is an unsupervised method for creating embeddings for genomic regions and region sets from a set of raw BED files. The program will first map all raw regions to a given universe (vocabulary) set. Then, it will construct sentences by concatenating regions from a BED file in random order. The generated sentences will be used for Region2Vec training using word2vec.

Usage

1. Prepare a set of bed files in src_folder. [Optional] If only a subset of files will be used, specify a list of those files as file_list. By default, the program will use all the files in the folder to train a Region2Vec model.
2. Prepare a universe file universe_file.
3. Create a token folder which will be used to store tokenized files dst_folder.
4. Run the following command

from geniml.tokenization import hard_tokenization
from geniml.region2vec import region2vec

src_folder = '/path/to/raw/bed/files'
dst_folder = '/path/to/tokenized_files'
universe_file = '/path/to/universe_file'

# must run tokenization first
status = hard_tokenization(src_folder, dst_folder, universe_file, 1e-9)

if status: # if hard_tokenization is successful, then run Region2Vec training
    save_dir = '/path/to/training/results'
    region2vec(dst_folder, save_dir, num_shufflings=1000)

For customized settings, please go and check the parameters used in main.py. For training a Region2Vec model, the parameters, init_lr, window_size, num_shufflings, embedding_dim, are frequently tuned in experiments.

For command line usage, type geniml region2vec --help for details. We give a simple usage below

geniml region2vec
  --token-folder /path/to/token/folder \
  --save-dir ./region2vec_model \
  --num-shuffle 10 \
  --embed-dim 100 \
  --context-len 50