SWAN-MPO: A Target-Calibrated Multi-Parameter Optimization Framework for Natural-Product Scaffold Triage in Oncology
Abstract Natural products occupy structural and physicochemical space that is poorly represented by optimization criteria developed for conventional small molecules. We developed SWAN-MPO, a multi-parameter framework that integrates predicted target binding, absorption/distribution/metabolism/excretion (ADME) properties, general safety, and disease-context liabiliy for natural-product scaffold triage. Receptor-specific docking calibration and bounded desirability transformations place heterogeneous evidence on a common decision scale before cross-domain integration. Applied to 59 Annona muricata compounds across nine oncology targets, SWAN-MPO substantially changed candidate prioritization by favoring balanced cross-domain profiles. External ChEMBL benchmarking revealed why this matters: binding almost perfectly distinguished active from inactive compounds (ROC-AUC 0.976) but was essentially uninformative for separating favorable from serious-warning profiles among already-active compounds (ROC-AUC 0.503), where ADME was more discriminative (ROC-AUC 0.620). On the combined activity/status endpoint, the integrated formulation achieved ROC-AUC 0.909. In a separate descriptor-compatible benchmark, SWAN-MPO significantly outperformed four adapted multiparameter and drug-likeness comparators in global discrimination. These results show that natural-product prioritization depends on integrating property domains that answer different screening questions rather than optimizing any single signal.
Authors
- Saanvii Senthilkumar (ORCID: https://orcid.org/0009-0000-2050-1070)
Publication Details
- Journal
- Research Square
- Published
- 2026-11-13
- DOI
- https://doi.org/10.21203/rs.3.rs-10863911/v1
- Primary Topic
- Computational Drug Discovery Methods
- Type
- preprint
Funders
- University of Texas MD Anderson Cancer Center