Target specific compound identification using a support vector machine

Area of Science:

• Computational chemistry
• Chemoinformatics
• Machine learning in drug discovery

Background:

• High-throughput screening (HTS) campaigns often begin with existing knowledge of active molecules.
• Known active compounds from related targets or literature can guide initial screening efforts.
• Virtual high-throughput screening (vHTS) is used when no initial active compounds are known, but often yields many false positives.

Purpose of the Study:

• To evaluate the effectiveness of a support vector machine (SVM) for classifying chemical compounds with unknown activity.
• To reduce the number of compounds requiring experimental testing in HTS campaigns.
• To improve the accuracy and speed of HTS hit list generation, especially when initial active compounds are scarce.

Main Methods:

• A supervised support vector machine (SVM) model was trained using atom pair (AP) two-dimensional topological descriptors.
• The SVM was trained on compounds from the MDL Drug Data Report (MDDR) known to be active against specific protein targets.
• The method was tested on five diverse biological targets: cyclooxygenase-2, dihydrofolate reductase, thrombin, HIV-reverse transcriptase, and estrogen receptor antagonists.

Main Results:

• The SVM model achieved high accuracy, with sensitivities exceeding 80% for all tested protein targets.
• Classification performance reached 100% for certain targets, demonstrating strong predictive power.
• The machine learning approach improved HTS speed by reducing the need for extensive docking procedures and enhanced the accuracy of hit lists.

Conclusions:

• Support vector machines utilizing atom pair descriptors provide an effective computational strategy for prioritizing compounds in HTS.
• This machine learning approach significantly enhances the efficiency and accuracy of drug discovery pipelines.
• The method offers a valuable alternative or complement to traditional vHTS, particularly when initial active compound data is limited.