Disease-Focused QSAR & Virtual Screening

We maintain a suite of 25+ disease-focused QSAR models that deliver rapid, decision-grade predictions of small-molecule efficacy. Each model is trained on manually curated, literature-validated datasets that integrate molecular interaction networks, pathway context, gene–disease associations, and metabolism/toxicity annotations. Models are rigorously validated (nested cross-validation and external test sets), calibrated for probability outputs, and deployed with applicability-domain and uncertainty estimates to guide risk-aware selection. Mechanistic interpretability (e.g., feature attributions) helps connect predictions to underlying biology.

These QSAR ensembles feed directly into our virtual screening pipeline: we filter ultra-large libraries, score compounds with disease models (e.g., for Alzheimer’s disease), flag liabilities (hERG/CYP/ADMET), and advance the best candidates to structure-based evaluation (e-pharmacophore, flexible docking with DRGSCROLL) and MD/free-energy refinement. The result is a fast, reproducible route from chemical space to experimentally testable leads linked to a specific disease mechanism.

Interested in collaborating or running a disease-specific screen? Contact: serdar.durdagi@bau.edu.tr