Developed Software & Tools (github)
Our research group actively develops state-of-the-art computational tools to accelerate and enhance the process of drug discovery and molecular modeling. These programs are designed to address critical challenges in pharmaceutical sciences by integrating physics-based simulations, advanced mathematical modeling, and artificial intelligence-driven algorithms.
Key features of our software platforms include:
· Innovative Docking Engines: Next-generation receptor-flexible docking programs that overcome the limitations of conventional docking methods.
· Pharmacophore Modeling & Virtual Screening Pipelines: Customized platforms enabling efficient hit identification and lead prioritization.
· AI-Integrated Predictive Models: Machine learning and deep learning frameworks tailored for QSAR analysis, toxicity prediction, and activity forecasting.
· Mathematical Scoring Functions: Novel approaches based on Banach space descriptors, Dirichlet series, and Morse theory to refine binding affinity predictions.
These tools are not only developed for our in-house research projects but are also made available to the wider scientific community through open-access web servers and collaborative platforms, aiming to foster innovation, reproducibility, and interdisciplinary collaboration in drug discovery.
DRGSCROLL is a receptor-flexible molecular docking engine developed by DurdağıLab (Lab4IND) at Bahçeşehir University. It targets cases where side-chain plasticity is essential (kinases, GPCRs, ion channels, cryptic/induced pockets).
Key ideas (vs. classic docking)
· Continuous χ-angle sampling: Replaces discrete rotamer libraries; side chains near the binding site are sampled on continuous torsional grids.
· Recurrent pose–receptor refinement: Poses are iteratively improved while resampling only the side chains that clash or are interaction-critical, keeping the rest of the protein stable.
· Clash-aware selection: Steric overlaps are detected/penalized early; non-feasible poses are filtered before scoring.
· Generation + clustering: Multiple “generations” of refinement; final poses are clustered by ligand heavy-atom RMSD and reported with representative structures.
When to use it
· Systems where side-chain rearrangements gate binding (e.g., kinases, GPCR microswitches, narrow or cryptic pockets).
· Early VS triage to reduce false negatives and false positives that rigid/classical docking generally misses.
Strengths & caveats
· Strengths: Physically realistic side-chain adaptation; avoids rotamer-combinatorics explosion; good early enrichment of “near-native” poses when judged by interaction/shape metrics.
· Caveats: More compute than rigid docking
Run: https://drgscroll.bau.edu.tr/
mol2drug — AI-powered molecular activity from images
mol2drug is a mobile application developed by DurdağıLab that uses state-of-the-art AI to analyze images of molecular structures and predict biological activities, potential targets/interactions, and drug-likeness—anytime, anywhere. It’s designed for researchers in pharma, biotech, and academia who need on-the-go decision support.
Key features
· AI-driven predictions from images: Upload camera shots of small molecules, sketches, or scans; the app recognizes and digitizes the molecule for instant analysis.
· Comprehensive activity reports: Summaries highlight predicted pharmacological properties (efficacy/toxicity context) and similarity to known drugs/targets.
· Continuous model updates: Models are regularly improved with new data and user feedback.
· Cross-disciplinary utility & mobile UX: Built for chemists and life-science teams, with a streamlined, accessible interface.
· Data protection: Encrypted transmission/storage for sensitive research data.