Welcome to CYP450model.org

CYP450model is a unified proteochemometric model for predicting drug inhibition of the five major cytochrome P450 enzyme isoforms (CYPs), namely CYP1A2, CYP2C9, CYP2C19, CYP2D6 and CYP3A4.

Background

CYPs are important drug and xenobiotic metabolizing enzymes, as well as important metabolizers of a large number of endogenous substances. A major problem in treatment of patients with drugs is so-called drug-drug interactions whereby one drug alters the function of another drug. A large part of drug-drug interactions occur via interactions with CYPs, which can lead to serious side effects. One mechanism for such drug-drug interactions is, for example, when drug A inhibits the metabolism of another drug B, particularly by virtue of drug A inhibiting the metabolism of drug B at a specific CYP enzyme. In drug discovery and development it is important to monitor such possibilities for drug-drug interactions at an early stage in the development process. This can be done by in vitro assays, which is a very expensive approach due to the many compounds that are commonly considered in the drug discovery process, and the many isoforms of CYPs available. An alternative is then to use an in silico prediction approach, to which end we here provide CYP450model, which is the first unified model for CYP-inhibition provided in one single model.

Proteochemometrics

Proteochemometics is a multi-target modeling technology, where the interactions of many targets with many interacting molecules are modelled simultaneously to create a unified model which spans the chemical space of all the targets over the entire chemical space of the library of interacting compounds covered by the model.

In short, proteochemometric modeling seeks to link structural information of molecular structures with their respective biological activity. It bears resemblance to QSAR in which a predictive model is constructed from a series of compounds toward a single protein target. By contrast, however, proteochemometrics considers information of multiple compounds and multiple protein targets in the development of its predictive model. This gives proteochemometrics the advantage that its unified models have better predictive ability and they provide more stable results when compared with single-target approaches, like QSAR. Moreover, proteochemometrics allows interpolation and even extrapolation towards new targets, not included in the model.

(For an introduction to proteochemometrics, see the book Introduction to Pharmaceutical Bioinformatics (ISBN: 978-91-979403-0-6) for which information is available here, and education on the technology which is available here).

CYP450model

CYP450model was developed based on the assay of 16 359 organic compounds towards their inhibition of the five CYPs using proteochemometrics modeling, all-in-all the dataset comprising 63 391 interactions.

The compounds were encoded by molecular signatures, which is essentially a canonical representation of the atomic environment up to a predefined height. The proteins were described by an alignment-independent approach where the composition and transition of amino acid properties in the primary sequence of proteins were considered, which is comprised of the following properties: 1) hydrophobicity, 2) normalized van der Waals volume, 3) polarity, 4) polarizability, 5) charge, 6) secondary structure, and 7) solvent accessibility. Predictive models were initially constructed using support vector machine, k-nearest neighbor, and random forest, and evaluated for their predictive performance. The model with support vector machine, using signatures of height 1, 2, and 3, was found to be most predictive, with internal AUC=0.923 and external AUC=0.940, and is the one provided here for CYP450model.

The model provides predictions of whether a drug is inhibitory or not on the CYP-enzymes. The cut-off level for a compound to be regarded as inhibitory is set to an IC50 of 10 µM or less for the compound.

Accessing CYP450model from Bioclipse

The proteochemometric CYP450model was created using Bioclipse, a free and open source software for the life sciences. In this way the CYP450model can be accessed directly from Bioclipse, as is illustrated below:

Shown in the above picture is for the left panel that CYP450model predicts that the drug quetiapine inhibits CYP2D6, in the middle panel that CYP450model predicts that the drug propranolol inhibits CYP1A2 and CYP2D6, and in the right panel that CYP450model predicts that the drug fluvoxamine inhibits all five CYPs.

Assessing the CYP-inhibitory activities of a compound takes only about 0.1 second. This means that you can edit a chemical structure in the Bioclipse editor and instantaneously get feedback for the new structure's CYP-inhibitory profile.

Installing and using CYP450model in Bioclipse

Binary downloads of Bioclipse for Windows, Linux, and Mac OS X can be accessed from http://www.bioclipse.net. The source code can be obtained from https://github.com/bioclipse.

How you install and run CYP450model is illustrated in this YouTube clip:

After you have downloaded and installed Bioclipse, you have to install the Decision Support feature of Bioclipse and then the proteochemometric CYP450model. You can then draw or import chemical structures and perform predictions of CYP450 inhibition, as illustrated in the clip.

Crediting the use of CYP450model

If you use CYP450model, please cite:

  • A unified proteochemometric model for predicting the inhibition of Cytochrome P450 isoforms.
    Maris Lapins, Apilak Worachartcheewan, Ola Spjuth, Valentin Georgiev, Virapong Prachayasittikul, Chanin Nantasenamat, Jarl E. S. Wikberg.
    PLoS One. 2013 Jun 17;8(6):e66566. doi: 10.1371/journal.pone.0066566.
  • Integrated decision support for assessing chemical liabilities.
    Spjuth O, Eklund M, Ahlberg Helgee E, Boyer S, Carlsson L.
    Chem Inf Model. 2011 Aug 22;51(8):1840-7. doi: 10.1021/ci200242c. Epub 2011 Aug 5.

Bioclipse Decision Support

The CYP450model is available from the Bioclipse Decision Support feature of Bioclipse, which is a general framework for executing multiple predictive models in parallel to assess chemical liabilities, such as to assess the biological, toxicological, chemical or physical properties of chemical compounds. Currently, besides the proteochemometric CYP450model, QSAR models are available for AMES mutagenicity, carcinogenicity, AhR, HeRG, and aquatic toxicity. The system can easily be extended with more models, new endpoints, and new data.

The chemical editor of Bioclipse allows you to edit chemical structures and rerun the models to get an updated result. As the predictions are very fast this can be done in real time, which allows you to try out different molecules and do structural modifications to optimize compound properties prior to synthesis. The feature is highly useful in drug discovery and lead optimization.

If you use Bioclipse Decision Support, please cite:

  • Integrated decision support for assessing chemical liabilities.
    Spjuth O, Eklund M, Ahlberg Helgee E, Boyer S, Carlsson L.
    J Chem Inf Model. 2011 Aug 22;51(8):1840-7. doi: 10.1021/ci200242c. Epub 2011 Aug 5.

Further Links

Contact

For inquiries contact Prof. Jarl Wikberg at Jarl.Wikberg@farmbio.uu.se