Datasets:

sarahantgan
/

5HT_Ki_Prediction

language:
  - en
license: cc-by-4.0
size_categories:
  - 100-1K
task_categories:
  - regression
pretty_name: 5-HT Ki Prediction Dataset
datasets:
  - sarahantgan/5HT_Ki_Prediction
tags:
  - bioactivity
  - cheminformatics
  - regression
  - serotonin
  - binding-affinity
dataset_info:
  features:
    - name: smiles
      dtype: string
    - name: ki
      dtype: float64
    - name: receptor
      dtype: string
    - name: source
      dtype: string
  splits:
    - name: train
      num_bytes: 5830493
      num_examples: 98678
  download_size: 780236
  dataset_size: 5830493
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

Serotonin Receptor (5-HT) Binding Affinity Prediction Dataset

This dataset was curated from the PDSP Ki Database to support training machine learning models that predict binding affinity (Ki in nM) of ligands to serotonin (5-HT) receptors.

Files Included

curated_ki_database.csv: Cleaned Ki dataset filtered for 5-HT targets
fingerprints_with_ki.csv: Molecules represented as 2048-bit Morgan (ECFP4) fingerprints
rf_model.pkl: Trained scikit-learn RandomForestRegressor model
test_predictions.csv: Predictions vs. actual Ki values on the test set
train_model.ipynb: Full Jupyter notebook with training code and evaluation

Modeling Approach

Molecules were standardized using MolVS.
SMILES strings were converted to 2048-bit Morgan fingerprints (ECFP4) using RDKit.
A Random Forest Regressor was trained using scikit-learn.
The model was evaluated on a held-out 20% test set.

Model Performance

R² Score: 0.257
RMSE: 4193.72 nM

📚 Citation & Source

Source: PDSP Ki Database
If reusing this dataset, please cite the PDSP Database appropriately.

👩‍💻 Author

Sara Hantgan
University of Michigan | BIOINF 595 Final Project
Winter 2025