A ranking method for the concurrent learning of compounds with various activity profiles

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• Developing effective virtual screening methods is crucial for multi-target drug design.
• Existing Support Vector Machine (SVM) classification techniques face challenges in handling compounds with diverse activity profiles across multiple targets.

Purpose of the Study:

• To introduce and evaluate a novel SVM-based ranking algorithm for concurrent learning of compounds with varying activity profiles.
• To enhance the prioritization of compounds in multi-target virtual screening scenarios.

Main Methods:

• A Support Vector Machine (SVM)-based ranking algorithm was developed.
• Compounds were specifically labeled to infer virtual screening models against multiple targets.
• The proposed algorithm was compared against state-of-the-art SVM classification techniques on three distinct chemical datasets.

Main Results:

• Ranking-based algorithms demonstrated superior performance in both single- and multi-target virtual screening.
• The SVM-based ranking method successfully ranked compounds with partially matching activity profiles higher than decoys.
• Performance improvements were observed compared to other multi-target SVM methods.

Conclusions:

• SVM-based ranking methods offer a valuable approach for virtual screening in multi-target drug design.
• These methods are particularly beneficial when dealing with compounds exhibiting diverse activity profiles.
• The approach is most useful when identifying numerous ligands with a perfectly matching activity profile is unlikely.