ENS-VS

ENS-VS applies ensemble machine learning to improve structure-based virtual screening by combining protein–ligand interaction energy terms and ligand structure vectors to predict compound activity.

Key Features:

• Ensemble Learning Framework: ENS-VS combines Support Vector Machine (SVM), Decision Tree, and Fisher Linear Discriminant classifiers in an ensemble to predict compound activity.
• Descriptor Utilization: The method employs protein–ligand interaction energy terms alongside ligand structure vectors as combination descriptors to characterize protein–ligand interactions.
• Enrichment Factor (EF) 1% Improvement: ENS-VS reported an EF 1% that is six times higher than AutoDock Vina in comparative evaluations.
• Benchmark Performance: On DUD-E ENS-VS achieved a mean EF 1% of 52.77 and an AUC of 0.982 (versus SIEVE-Score mean EF 1% 42.64 and AUC 0.912), and on DEKOIS achieved a mean EF 1% of 29.73 and an AUC of 0.793 (versus SIEVE-Score mean EF 1% 25.56 and AUC 0.765).

Scientific Applications:

• Structure-based Virtual Screening (SBVS): ENS-VS can be applied as an alternative and complementary computational approach to experimental high-throughput screening (HTS) for prioritizing compounds in SBVS campaigns.
• Computational Drug Discovery: The method supports identification and prioritization of promising small-molecule candidates in structure-based drug design workflows.

Methodology:

ENS-VS trains an ensemble of SVM, Decision Tree, and Fisher Linear Discriminant classifiers using combination descriptors composed of protein–ligand interaction energy terms and ligand structure vectors and evaluates performance using EF 1% and AUC on DUD-E and DEKOIS benchmarks.