PharmaCore: The Automatic Generation of 3D Structure-Based Pharmacophore Models from Protein/Ligand Complexes

In this work, we present PharmaCore: a new, completely automatic workflow aimed at generating three-dimensional (3D) structure-based pharmacophore models toward any target of interest. The proposed approach relies on using cocrystallized ligands to create the input files for generating the pharmacophore hypotheses, integrating not only the three-dimensional structural information on the ligand but also data concerning the binding mode of these molecules put in the protein cavity. We developed a Python library that, starting from the specific UniProt ID of the protein under investigation as the only element that requires user intervention, subsequently collects and aligns the corresponding structures bearing a known ligand in a fully automated fashion, bringing them all into the same coordinate system. The protocol includes a final phase in which the aligned small molecules are used to produce the pharmacophore hypotheses directly onto the protein structure using a specific software, e.g., Phase (Schrödinger LLC). To validate the entire procedure and highlight the possible applications in the field of drug discovery and repositioning, we first generated pharmacophores for soluble epoxide hydrolase (sEH) and compared with already-published ones. Then, we reproduced the binding profile of a reported selective binder of ATAD2 bromodomain (AM879), testing it against a panel of 1741 pharmacophores related to 16 epigenetic proteins and automatically generated with PharmaCore, finally disclosing putative unprecedented off-targets. The computational predictions were successfully validated with AlphaScreen assays, highlighting the applicability of the proposed workflow in drug discovery and repositioning. Finally, the process was also validated on tankyrase 2 and SARS-CoV-2 MPro, confirming the robustness of PharmaCore.本研究提出了PharmaCore：一种全新的全自动工作流，旨在为任意目标靶点生成基于三维（3D）结构的药效团模型（pharmacophore model）。所提出的方法依托共结晶配体构建生成药效团假说（pharmacophore hypothesis）所需的输入文件，不仅整合配体的三维结构信息，还涵盖了这些分子在蛋白空腔中的结合模式相关数据。我们开发了一款Python库，仅需用户提供所研究蛋白的特定UniProt编号（UniProt ID）作为唯一人工干预项，即可全自动完成后续步骤：收集并比对携带已知配体的对应蛋白结构，将所有结构统一至同一坐标系下。该流程包含最终阶段：利用比对后的小分子，借助专用软件（如薛定谔有限责任公司（Schrödinger LLC）的Phase软件（Phase））直接在蛋白结构上生成药效团假说。为验证整套流程的有效性并凸显其在药物发现与药物重定位领域的应用潜力，我们首先针对可溶性环氧化物水解酶（soluble epoxide hydrolase, sEH）生成了药效团模型，并与已发表的同类模型进行对比。随后，我们复现了已有报道的ATAD2溴结构域（ATAD2 bromodomain）选择性结合剂AM879的结合谱，将其与由PharmaCore自动生成的、对应16种表观遗传蛋白（epigenetic proteins）的1741个药效团组成的集合进行测试，最终揭示了此前未被报道的潜在脱靶靶点。该计算预测通过AlphaScreen实验（AlphaScreen assay）得到了成功验证，进一步证实了所提工作流在药物发现与重定位领域的应用价值。最后，我们分别在端锚聚合酶2（tankyrase 2）与SARS-CoV-2主蛋白酶（MPro）上验证了该流程，证实了PharmaCore的稳健性。