Salomon-Ferrer, R. | Search bioRxiv

bioRxiv2026

Discovery of Selective Small-Molecule Ligands of SV2C by AI-Enhanced Virtual Screening and Experimental Validation

Synaptic vesicle glycoprotein 2C (SV2C) is a vesicular protein enriched in dopaminergic neurons of the basal ganglia that modulates dopamine storage and release, and its disruption is implicated in Parkinson's disease (PD). Despite strong genetic and pathological links to PD, there are no selective small-molecule probes for SV2C. Here, we describe an AI-enhanced virtual screening (VS) and experimental campaign that identified multiple novel chemotypes with low-micromolar affinity and marked selectivity for SV2C over SV2A and SV2B, starting from a large, general-purpose commercial library. Because no full-length high-resolution SV2C structure was available, we built a homology model using SV2A cryo-EM structures as templates and characterized its conformational landscape by molecular dynamics (MD) and Gaussian accelerated MD (GaMD) simulations in apo form and in complex with known SV2 ligands (plosaracetam, levetiracetam, brivaracetam, and padsevonil). A convolutional neural network-based scoring function (CNN_VS), retrospectively validated on a manually curated 39-ligand SV2A benchmark (r = 0.72 vs experimental pIC50), was then applied in a multi-stage funnel to 5.96 million Mcule in-stock compounds, which were sequentially filtered to 3.19 million CNS-relevant molecules before docking and rescoring. From 94 VS-prioritized candidates, 71 compounds were experimentally profiled in an orthogonal primary assay cascade combining a thermal shift assay (TSA) with a [3H]-padsevonil scintillation proximity assay (SPA), followed by Ki determination and isoform selectivity profiling for key hits. This campaign yielded 22 active molecules (31% hit rate) that naturally segregated into two categories: compounds that showed primary site competition, and compounds that did not show primary site competition with [3H]-padsevonil. A subset of competitor compounds also showed thermostabilization activity. Among these, compounds 36 and 56 emerged as particularly attractive leads, with Ki values of 24.6 uM and 3.25 uM at SV2C, respectively, and greater than 10-fold selectivity versus SV2A; compound 56 also maintained approximately 12-fold selectivity relative to SV2B. A complementary subset of SV2C-selective hits behaved as padsevonil-site competitors, providing a lead set that will serve as a template for functional characterization and future drug development for conditions that affect dopaminergic signaling. Docking analysis suggests a common binding mode anchored by conserved tryptophan residues in the SV2 pocket, a prediction independently confirmed by an unpublished SV2A-plosaracetam cryo-EM structure showing 0.76 Angstrom binding-site C-alpha RMSD relative to the SV2C model and complete conservation of the tryptophan cage. Subtle differences in the luminal domain and transmembrane region point to the structural determinants underlying isoform selectivity. Collectively, these results demonstrate that an AI-driven VS pipeline, tightly integrated with medium-throughput biophysical assays, can deliver selective SV2C binders from a general chemical library on a structurally under-characterized membrane target. The identified hits provide multiple starting points for hit-to-lead optimization and tools for probing SV2C biology and its role in PD.

neuroscience↗

bioRxiv2025

SAIR: Enabling deep learning for protein-ligand lnteractions with a synthetic structural dataset

AO_SCPLOWBSTRACTC_SCPLOWAccurate prediction of protein-ligand binding affinities remains a cornerstone problem in drug discovery. While binding affinity is inherently dictated by the 3D structure and dynamics of protein-ligand complexes, current deep learning approaches are limited by the lack of high-quality experimental structures with annotated binding affinities. To address this limitation, we introduce the Struc-turally Augmented IC50 Repository (SAIR), the largest publicly available dataset of protein-ligand 3D structures with associated activity data. The dataset com-prises 5, 244, 285 structures across 1, 048, 857 unique protein-ligand systems, cu-rated from the ChEMBL and BindingDB databases, which were then computa-tionally folded using the Boltz-1x model. We provide a comprehensive charac-terization of the dataset, including distributional statistics of proteins and ligands, and evaluate the structural fidelity of the folded complexes using PoseBusters. Our analysis reveals that approximately 3% of structures exhibit physical anoma-lies, predominantly related to internal energy violations. As an initial demon-stration, we benchmark several binding affinity prediction methods, including empirical scoring functions (Vina, Vinardo), a 3D convolutional neural network (Onionnet-2), and a graph neural network (AEV-PLIG). While machine learning-based models consistently outperform traditional scoring function methods, nei-ther exhibit a high correlation with ground truth affinities, highlighting the need for models specifically fine-tuned to synthetic structure distributions. This work provides a foundation for developing and evaluating next-generation structure and binding-affinity prediction models and offers insights into the structural and phys-ical underpinnings of protein-ligand interactions. The dataset can be found at https://www.sandboxaq.com/sair.

biochemistry↗