SilicoSAR: An integrated machine-learning QSAR module within the SilicoXplore platform for QSAR-based virtual screening, demonstrated by the discovery of natural-product EGFR inhibitors.

Abstract

Quantitative structure-activity relationship (QSAR) modelling is central to computer-aided drug discovery, yet many existing QSAR tools operate as isolated descriptor-generation or single-algorithm utilities that require programming expertise and manual transfer of outputs into separate docking and simulation programs. We present SilicoSAR, an integrated, graphical machine-learning QSAR module within the SilicoXplore drug-discovery platform that unifies data curation, descriptor calculation, feature selection, multi-algorithm model building and validation, applicability-domain assessment, and large-scale virtual screening in a single reproducible pipeline. Its capabilities are demonstrated through an end-to-end case study targeting the Epidermal Growth Factor Receptor (EGFR), a clinically important anticancer target. A curated dataset of approximately 10,000 EGFR inhibitors from BindingDB was used to build and validate an XGBoost QSAR model, which then screened approximately 36,000 natural compounds from NPAtlas, followed by ADMET filtering, docking, pharmacophore-guided prioritisation, molecular dynamics (MD) simulations, MM-GBSA binding free-energy calculations, and free-energy perturbation (FEP) analysis, all within SilicoXplore. The model achieved R² values of 0.92, 0.58, and 0.61 for training, test, and external validation sets, respectively. Screening identified approximately 16,000 predicted-active compounds, reduced to 2035 after ADMET filtering and 280 after docking. Twenty structurally diverse compounds were prioritised for MD, and eight leads showed stable binding over 200 ns simulations. MM-GBSA and FEP analyses highlighted NPA004311 and NPA004681 as the most promising candidates. This case study confirms that SilicoSAR generates statistically reliable QSAR models and drives them through an integrated structure-based pipeline, demonstrating the value of a unified, accessible QSAR-to-simulation workflow for accelerated, natural-product-based drug discovery.