Main page

About pkCSM

pkCSM is a machine-learning platform to predict small-molecule pharmacokinetic properties, which relies on distance/pharmacophore patterns encoded as graph-based signatures. The platform is composed of 28 regression and classification models, trained and tested on different experimental data sets encompassing a diverse and complementary set of ADMET descriptors as follows:

• Absorption: Caco-2 permeability, water solubility, intestinal absorption (human), P-glycoprotein substrate/inhibitor and skin permeability;


• Distribution: Volume of distribution (human), fraction unbound (human) BBB and CNS permeability;


• Metabolism: Cytochrome P450 inhibitor and substrate;


• Excretion: Renal OCT2 substrate and Total Clearance;


• Toxicity: Rat LD50, AMES toxicity, T.Pyriformis toxicity, Minnow toxicity, Maximum Tolerated Dose, Oral Rat Chronic toxicity, Hepatotoxicity and Skin sensitisation.

More information about the predictive models and how to interpret the pkCSM predicitons can be accessed via the Theory menu (1).

Submission page

How to run a prediction

To run a prediction:

• Click on "Prediction" (1) to open the submission page.


• Provide either an input file with a list of molecules (2) in SMILES format (up to a maximum of 100 molecules) or supply a single SMILES string (3). Users are advised to use Canonical SMILES. Syntax non-compliant molecules will be ignored.


• The next step is to choose the prediction mode (4). Users can choose between the main ADMET property classes (Absorption, Distribution, Metabolism, Excretion and Toxicity) by clicking on its corresponding button or run a systematic evaluation of all predictive models.

Results page

Results

For a single molecule prediction, your results can be displayed as follows:

• A depiction of the uploaded molecule will be shown in (1). Make sure the depiction is what you are expecting for your molecule.


• A list of molecule properties will be calculated and shown in (2).


• The prediction will be displayed in a tabular format as presented in (3). The information shown include the ADMET property being predicted, model name, the actual predicted value and whether the prediction is numeric (for regression models, including the unit of the predicted value) or categoric (for classification models).

Results for multiple molecules will be shown in a tabular format, whithout molecule depiction, which can be downloaded as a CSV file.

Contact page

Getting in touch

In case you experience any trouble using pkCSM or have any suggestions or comments, please do not hesitate in contacting us (1) either via e-mail or through the online form.