API and Docs

---

API

The API for the model can be accessed via this link.

There are also user interfaces PredGUI and PredGUIMM where the API is used.

Model Details

---

Overview

• This models predicts water–octanol distribution coefficient (logD) for chemical compounds: It is a proxy for the lipophilicity which is a major determinant of drug properties and overall suitability of drug candidates.


• The model is based on data for 1.6 million compounds from the ChEMBL database with available predicted values for ACD/logD. For modeling, a support-vector machine with a linear kernel using conformal prediction methodology was used, outputting prediction intervals at a specified confidence level. Features for chemical structures were Signature Molecule Descriptor calculated with Chemistry Development Kit.

Version

Owners

• Ola Spjuth

References

• Lapins M, Arvidsson S, Lampa S, Berg A, Schaal W, Alvarsson J, Spjuth O. A confidence predictor for logD using conformal regression and a support-vector machine. Journal of Cheminformatics. 10, 18 (2018). DOI: 10.1186/s13321-018-0271-1

Model Architecture

• The model is based on data for 1.6 million compounds from the ChEMBL database with available predicted values for ACD/logD. For modeling, a support-vector machine with a linear kernel using conformal prediction methodology was used, outputting prediction intervals at a specified confidence level.


• For more details, see the scientific paper A confidence predictor for logD using conformal regression and a support-vector machine - Journal of Cheminformatics

Input Format

• A chemical structure in SMILES or MOL format.
• A confidence level for predictions.
• [optional] Size of image for highlighting predictions in chemical structure.

Output Format

• A prediction interval for cpLogD at the selected confidence level.
• Image for highlighting the relevant atoms in chemical structure that contributed the most to the prediction.

Considerations

Use Cases

• For novel chemical structures in drug discovery projects, this model can give a prediction of its lipophilicity






• Examples of molecule gradients for the prediction of cpLogD. Shown are gradients for four compounds indicated by arrows in Fig. 3. Upper row: atenolol (logD=−1.82) and sotalol (logD=−1.52). Lower row: tolnaftate (logD=5.4) and amiodarone (logD=6.1)

Limitations

• The model is trained in values calculated suing the ACD_LogP algorithm, not measured values.

Accuracy

---

The model shows a predictive ability of Q2=0.973 and with the best performing nonconformity measure having median prediction interval of ± 0.39 log units at 80% confidence and ± 0.60 log units at 90% confidence. Plotted are acd_logd values (x-axis) versus the predicted logD values (y-axis) for 100,000 test set compounds. The root mean square error of prediction (RMSEP) is 0.41 log units.