DFG Conformational Benchmark — Structural Data Archive (Predicting Ligand-Induced Conformational Selection)

Companion dataset for the manuscript Predicting Ligand-Induced Conformational Selection: Co-Folding, Generative Docking, and the Pharmacophore Understanding Problem. This deposit contains every input crystal structure, pseudo-apo input, predicted protein-ligand complex, raw pipeline output, PoseView 2D interaction diagram, and pre-built PyMOL alignment session referenced in the manuscript. Five deep-learning docking/co-folding methods are benchmarked (Boltz-2, FlowDock, DynamicBind, DiffDock, Gnina) across kinases with DFG-in/DFG-out switching (BRAF V600E, ABL, p38α, c-Kit, VEGFR2), nuclear receptor ERα (helix 12 repositioning), and the β2AR GPCR reference. The sorafenib chemical deconstruction fragments (S0..S7) used to test whether co-folding captures pharmacophore logic are included under deconstruction/. Analysis scripts and numeric output tables are available at https://github.com/CONNECTS-SCV/dfg-conformational-benchmark. See README.md at the root of this archive for the full directory layout and reproduction instructions.