ADMET-AI

Train Chemprop ADMET predictors using the multitask datasets via scripts/train_tdc_admet_all.py. The script will use a GPU by default if one is available.

python scripts/train_tdc_admet_all.py \
    --data_dir data/tdc_admet_all_multitask \
    --save_dir models/tdc_admet_all_multitask \
    --model_type chemprop

Evaluate Chemprop ADMET predictors trained on the TDC ADMET Benchmark Group data using scripts/evaluate_tdc_admet_all_multitask.py.

Environment Requirement: The PyTDC environment should be used for this command.

python scripts/evaluate_tdc_admet_all_multitask.py \
    --data_dir data/tdc_admet_all_multitask \
    --preds_dir models/tdc_admet_all_multitask/chemprop \
    --save_dir results/tdc_admet_all_multitask

Generate multitask datasets for regression and classification from the TDC ADMET datasets using scripts/merge_tdc_admet_multitask.py.

python scripts/merge_tdc_admet_multitask.py \
    --data_dir data/tdc_admet_all \
    --save_dir data/tdc_admet_all_multitask

Extract approved drugs from a drugbank.xml file (v5.1.14) to create a comparison set using scripts/get_drugbank_approved.py.

python scripts/get_drugbank_approved.py \
    --data_path data/drugbank/drugbank.xml \
    --save_path data/drugbank/drugbank_approved.csv

Ensure all CSV files in your dataset contain SMILES that result in valid RDKit molecules by running scripts/filter_valid_molecules.py.

Environment Requirement: Use the general ADMET-AI Python environment with an up-to-date RDKit.

python scripts/filter_valid_molecules.py \
    --data_dir data/tdc_admet_all \
    --smiles_column smiles

You can install ADMET-AI using pip. It is recommended to use a conda environment. By default, the installation includes only dependencies for predictions (CLI and Python API). To include dependencies for the web server, use the [web] extra.

Standard Installation

# Optional: Create and activate a conda environment
conda create -y -n admet_ai python=3.14
conda activate admet_ai

# Install via pip
pip install admet-ai

Local Installation from Source

git clone https://github.com/swansonk14/admet_ai.git
cd admet_ai
pip install -e .

Install Web Server Dependencies

pip install admet-ai[web]

Troubleshooting

If you encounter ImportError: libXrender.so.1: cannot open shared object file: No such file or directory, run:

conda install -c conda-forge xorg-libxrender

To download all TDC ADME and Tox datasets for training, use the scripts/prepare_tdc_admet_all.py script.

Environment Requirement: This specific command requires a Python 3.10 environment containing PyTDC==1.1.15 and typed-argument-parser==1.11.0. Note that PyTDC is incompatible with the standard ADMET-AI environment requirements.

python scripts/prepare_tdc_admet_all.py \
    --save_dir data/tdc_admet_all \
    --skip_datasets herg_central hERG_Karim ToxCast

You can generate DrugBank reference plots and radial summary plots locally using provided scripts. Both scripts require a CSV file containing ADMET-AI predictions as input.

• DrugBank reference plots: Use scripts/plot_drugbank_reference.py
• Radial summary plots: Use scripts/plot_radial_summaries.py

You can integrate ADMET predictions into your Python workflows using the ADMETModel class from the admet_ai module.

Behavior:

• If you pass a single SMILES string, predict() returns a dictionary mapping property names to values.
• If you pass a list of SMILES strings, predict() returns a Pandas DataFrame where the index is the SMILES and the columns are the properties.

from admet_ai import ADMETModel

model = ADMETModel()
# Returns a dictionary
preds = model.predict(smiles="O(c1ccc(cc1)CCOC)CC(O)CNC(C)C")

# If passing a list, returns a Pandas DataFrame
# preds = model.predict(smiles=["SMILES1", "SMILES2"])

Use plot_drugbank_reference to generate a 2D scatter plot that overlays new molecule predictions (preds_df) on top of a DrugBank reference dataset (drugbank_df). The plot includes marginal histograms for the selected properties.

Parameters:

• preds_df (pd.DataFrame): Predictions for the new molecules.
• drugbank_df (pd.DataFrame): The DrugBank reference set.
• x_property_name (str, optional): Name of the property for the x-axis. Defaults to `

Use the admet_predict CLI tool to make ADMET predictions and compute physicochemical properties on DrugBank approved drugs using a Chemprop multitask predictor.

admet_predict \
    --data_path data/drugbank/drugbank_approved.csv \
    --save_path admet_ai/resources/data/drugbank_approved.csv \
    --models_dir models/tdc_admet_all_multitask/chemprop \
    --smiles_column smiles \
    --drugbank_path None

Use the admet_predict command to generate ADMET predictions from a file containing SMILES strings.

Arguments:

• --data_path: Path to the input file (e.g., a CSV).
• --save_path: Path where the prediction results will be saved.
• --smiles_column: The name of the column in the input file containing the SMILES strings.

admet_predict \
    --data_path data.csv \
    --save_path preds.csv \
    --smiles_column smiles