Cysteine Structure Database

The [Cysteine Structure Database] is a dataset compiled of strucutral data for 6515 cysteine sites in hundreds of proteins. This dataset was published in Cell and is also available at the official DrugMap Github repo.

For each cysteine site, this database includes numerical values for Solvent Accessible Surface Area (SASA), Cysteine Depth, etc. Additionally, each cysteine site has a probe engagement score derived from isotopic tandem orthogonal proteolysis-activity-based protein profiling (isoTOP-ABPP) that is represented as True or False in this dataset for three probes: KB02, KB03, KB05.

Quickstart Usage

Load a Dataset in Python

Each subset can be loaded into python using the Huggingface datasets library. Install the datasets library

$ pip install datasets

then, in Python, load the datasets library

>>> import datasets

and load one of the Cysteine Structure Database datasets, e.g.,

>>> KB03_data = datasets.load_dataset('ymanasa2000/DrugMap_Ligandability', name='KB03')
Downloading readme: 100%|████████████████████████| 4.04k/4.04k [00:00<00:00, 281kB/s]
Downloading data: 100%|████████████████████████| 30.5k/30.5k [00:00<00:00, 470kB/s]
Downloading data: 100%|████████████████████████| 218k/218k [00:00<00:00, 2.55MB/s]
Downloading data: 100%|████████████████████████| 54.8k/54.8k [00:00<00:00, 915kB/s]
Generating test split: 100%|████████████████████████| 143/0 [00:00<00:00, 4939.35 examples/s]
Generating train split: 100%|████████████████████████| 1029/0 [00:00<00:00, 40692.99 examples/s]
Generating val split: 100%|████████████████████████| 258/0 [00:00<00:00, 22803.78 examples/s]

Then, inspect the loaded dataset

>>> KB03_data
DatasetDict({
  test: Dataset({
      features: ['Unnamed: 0', 'site', 'depth', 'absolute_sasa', 'hse_up', 'hse_down', 'coord_number', 'rsa', 'h_nho1', 'h_ohn1', 'h_nho2', 'h_ohn2', 'tco', 'kappa', 'alpha', 'phi', 'psi', 'pocket', 'interface', 'basic', 'acidic', 'polar', 'cysteine', 'structural', 'aliphatic', 'aromatic', 'KB03', 'struct_motif_B', 'struct_motif_E', 'struct_motif_G', 'struct_motif_H', 'struct_motif_I', 'struct_motif_P', 'struct_motif_S', 'struct_motif_T'],
      num_rows: 143
  })
  train: Dataset({
      features: ['Unnamed: 0', 'site', 'depth', 'absolute_sasa', 'hse_up', 'hse_down', 'coord_number', 'rsa', 'h_nho1', 'h_ohn1', 'h_nho2', 'h_ohn2', 'tco', 'kappa', 'alpha', 'phi', 'psi', 'pocket', 'interface', 'basic', 'acidic', 'polar', 'cysteine', 'structural', 'aliphatic', 'aromatic', 'KB03', 'struct_motif_B', 'struct_motif_E', 'struct_motif_G', 'struct_motif_H', 'struct_motif_I', 'struct_motif_P', 'struct_motif_S', 'struct_motif_T'],
      num_rows: 1029
  })
  val: Dataset({
      features: ['Unnamed: 0', 'site', 'depth', 'absolute_sasa', 'hse_up', 'hse_down', 'coord_number', 'rsa', 'h_nho1', 'h_ohn1', 'h_nho2', 'h_ohn2', 'tco', 'kappa', 'alpha', 'phi', 'psi', 'pocket', 'interface', 'basic', 'acidic', 'polar', 'cysteine', 'structural', 'aliphatic', 'aromatic', 'KB03', 'struct_motif_B', 'struct_motif_E', 'struct_motif_G', 'struct_motif_H', 'struct_motif_I', 'struct_motif_P', 'struct_motif_S', 'struct_motif_T'],
      num_rows: 258
  })

})

Use a Dataset to Train a Model

One way to use the dataset is by training a Baseline Random Forest Classifier to predict intereaction of a cysteine with one of the three probes (KB02, KB03, KB05). In this example, we will train and test on KB03 data.

First, install scikit-learn

>>> pip install scikit-learn

then load, split, featurize, fit and evaluate the Random Forest model

from sklearn.ensemble import RandomForestClassifier
import pandas as pd

KB03_data = datasets.load_dataset('ymanasa2000/DrugMap_Ligandability', name='KB03')

# split into train and test
KB03_train = KB03_data['train']
KB03_test = KB03_data['test']

train_set = pd.DataFrame(KB03_train)
test_set = pd.DataFrame(KB03_test)

# featurize
X_train = train_set.drop(columns=['site', 'KB03'])
y_train = train_set['KB03']

X_test = test_set.drop(columns=['site', 'KB03'])
y_test = test_set['KB03']

# fit
model_1 = RandomForestClassifier()
model_1.fit(X_train, y_train)

# evaluate 
print(model_1.score(X_test, y_test)) # output: 0.5944

About the Cysteine Structure Database

Features of the DB

This DB features a csv with structural data for ~6,500 bindable cysteines in hundreds of protein active sites. Each cysteine has structural data such as, numerical values for Solvent Accessible Surface Area (SASA), Cysteine Depth, etc. Additionally, this DB contains probe read-outs from an experiment described in Takahashi_et_al_2024. They integrated the isotopic tandem orthogonal proteolysis-activity-based protein profiling (isoTOP-ABPP) platform with tandem mass tag (TMT)-based mass spectrometry quantification (iso-TMT) to measure cysteine reactivity. In this approach, cell lysates are first treated with cysteine-reactive “scout” compounds or vehicle control, allowing reactive cysteines a chance to form covalent adducts, and then this is followed by a chase with a pan-cysteine-reactive probe (iodoacetamide-desthiobiotin DBIA), which reacts with all remaining free cysteine thiolate groups. Crucially, cysteines that reacted with the scout compound will escape being tagged by DBIA. Ligandable cysteines are defined as those that are engaged (ε-value) >60% by cysteine-reactive compounds.

Data splits

The authors of this dataset suggested using a Stratified Split via the train_test_split() method which was used to produce the datasets in this Hugging Face DB.

Citation

Please use the following citation in any publication using our Cysteine Structure Dataset:

@article{
    Takahashi_et_al_2024, 
    author={Takahashi, Chong, Harrison, Bar-Peled, et al},
    doi={10.1016/j.cell.2024.03.027},
    journal={Cell},
    number={10},
    month={May}
    title={DrugMap: A quantitative pan-cancer analysis of Cysteine ligandability},
    volume={187},
    year={2024}
    url = {https://www.biorxiv.org/content/10.1101/2023.10.20.563287v1}
  }

Datasets:

ymanasa2000
/

DrugMap_Ligandability

You need to agree to share your contact information to access this dataset