An improved approach for predicting drug-target interaction: proteochemometrics to molecular docking

Area of Science:

• Computational chemistry
• Cheminformatics
• Pharmacology

Background:

• Proteochemometric (PCM) methods predict drug-target interactions (DTI) using descriptors of both drugs and targets.
• Key challenges in PCM include generating reliable non-interacting datasets and defining model applicability domains (AD).

Purpose of the Study:

• To improve traditional PCM modeling by developing novel methods for negative dataset generation and fingerprint-based AD analysis.
• To evaluate various descriptors and classifiers for DTI prediction performance.
• To demonstrate the practical utility of the developed models by predicting targets for natural anticancer drugs.

Main Methods:

• Developed novel methodologies for reliable negative dataset generation and fingerprint-based AD analysis within PCM.
• Evaluated diverse descriptors and classifiers, including Random Forest and Support Vector Machine.
• Employed reverse molecular docking to quantify molecular recognition interactions for predicted drug-target pairs.

Main Results:

• Random Forest and Support Vector Machine models achieved high accuracies (>98% cross-validation, >89% external validation).
• Sequence-based protein descriptors were found effective, with negligible impact from structure-based descriptors.
• Predicted targets for natural anticancer drugs largely aligned with known anticancer therapies, with 30 DTIs confirmed in the ChEMBL database.

Conclusions:

• The improved PCM methodology, combined with molecular docking, is effective for elucidating drug mechanisms of action.
• This approach can successfully identify potential drug targets and facilitate drug repositioning for new therapeutic applications.
• The study highlights the utility of sequence-based descriptors and robust AD analysis in PCM modeling.