metadata
dataset_info:
features:
- name: id
dtype: string
- name: seqs
sequence: string
- name: contacts
sequence:
sequence: uint16
splits:
- name: train
num_bytes: 10397872068.85042
num_examples: 142277
- name: valid
num_bytes: 73081890.0373948
num_examples: 1000
- name: test
num_bytes: 73081890.0373948
num_examples: 1000
download_size: 4574628106
dataset_size: 10544035848.92521
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: valid
path: data/valid-*
- split: test
path: data/test-*
All PDB proteins as of 5/24/24. Contacts recorded in "COO" format, where a contact gets a tuple (i,j) if the ith and jith amino are within 8 angstrom. In huggingface datasets format is a nonredundant selection. The entire 200 thousand are uplaoded as pickle files.