metadata
dataset_info:
features:
- name: IdA
dtype: string
- name: IdB
dtype: string
- name: labels
dtype: float64
- name: mechanism
dtype: string
- name: effect
dtype: string
- name: score
dtype: float64
- name: sentence
dtype: string
- name: signor_id
dtype: string
splits:
- name: train
num_bytes: 12875889
num_examples: 31474
download_size: 4977813
dataset_size: 12875889
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
Downloaded from SIGNOR 4.0 on 2/4/2026: https://signor.uniroma2.it/downloads.php
Used the Complete Dataset for Homo sapiens
- Filtered to entries where TYPEA=='protein' and TYPEB=='protein'
- Filtered to entries where DIRECT=='YES'
- Dropped entries where EFFECT=='uknown'
- Dropped a single entry where EFFECT=='form complex'
- Filtered to only include the shown columns (renamed)
- Dropped duplicates
Assigned a label of 2 to entries whose EFFECT was 'binding', as well as entries for which the invert appeared (If IDA=X and IDB=Y, when the inverse exists in the dataset, those get a label of 2)
The rest of the entries received a label of 1. The IDs for those entries were then inverted and added to the dataset. The inverted pairs received a label of 0.