DRGSCROLL: Achieving Full Side-Chain Flexibility in Docking Simulations through Genetic Algorithm Framework

Abstract

Structure-based docking often assumes a rigid receptor, obscuring induced-fit side-chain rearrangements that govern affinity and selectivity. We introduce DRGSCROLL, an open access docking platform and web server that jointly optimizes ligand pose and continuous receptor side-chain {chi} angles within a genetic-algorithm (GA) optimizer. DRGSCROLL seeds broad {chi}-angle populations, evaluates candidates with a dual-objective fitness that rewards low interaction energy profile while penalizing steric clashes, and uses per-residue crossover plus stochastic mutation to maintain physically realistic {chi} sets. To favor exploration over premature local minima, per-iteration minimization is deliberately omitted; final poses are selected by clash-free filtering and RMSD-based clustering. Across PDBbind protein-ligand complexes, DRGSCROLL showed generation-wise decreases in clash counts and improved docking scores, indicating convergence to sterically viable, low-energy pocket conformations rarely accessed by rigid-receptor protocols. Relative to known publicly available and commercial docking programs such as Vina, Glide, and IFD; DRGSCROLL sampled more favorable energy distributions with lower median scores, consistent with superior induced-fit capture. In prospective virtual-screening evaluations, DRGSCROLL enhanced actives versus inactives discrimination for RET kinase and PARP1 targets, while maintaining balanced precision-recall trade-offs. Additionally, DRGSCROLLs endurance in differentiating actives from property-matched decoys was validated by benchmarking against a decoy database (Directory of Useful Decoys, DUDE), which showed improved AUC, recall, and enrichment performance in comparison to Vina GPU 2.1 under the same screening settings. By embedding continuous side-chain flexibility directly into the search, rather than relying on post hoc rotamer tweaks or heavy local minimization, DRGSCROLL addresses key combinatorial and feasibility bottlenecks in flexible docking. The method provides a scalable, physics-grounded route to adaptive receptor-ligand modeling that improves pose accuracy and early enrichment for flexible targets, from fragment screening to triage of synthetically ready or AI-generated small molecule libraries. Availability: platform page, https://www.drgscroll.com/; academic/nonprofit web server, https://drgscroll.bau.edu.tr/