肝X受体β亚型选择性配体设计的深度学习方法研究

The liver X receptor (LXR) has two subtypes of α and β. Due to the small difference in the ligand binding domain, its structure-based selective ligand design is extremely difficult. In the previous study, we analyzed the structural data of a large number of LXR ligands using a Bayesian learning machine and found a new scaffold of LXR ligands. It is also found that the attentional mechanism of deep learning can identify the structural features of the targeting ligand. Therefore, we believe that based on the new scaffold of LXR ligands, the attentional mechanism of deep learning can be used to identify the structural specificity of selective ligands and to find highly selective LXRβ ligands. In order to verify this hypothesis, this project intends to study the following contents: (1) establish a library of known LXRβ and LXRα ligands; (2) establish a deep learning model of attention mechanism, and discover the structural features of LXRβ selective ligands; (3) Design, synthesize and screen LXRβ selective ligands based on the new scaffold of LXR ligand using the established deep learning model; (4) Evaluate the anti-glioblastoma activity of LXRβ selective agonists. The successful implementation of this project not only finds new methods for LXR selective ligand design, but also lays a new foundation for the final discovery of the selective mechanism of LXRβ and LXRα ligands.

肝X受体（LXR）有α与β两个亚型，由于配体结合域差别微小，基于结构的选择性配体设计困难。前期研究中，我们用贝叶斯学习机分析了大量LXR配体数据，发现了LXR配体新骨架，并且发现深度学习的注意力机制可以识别靶向配体的结构特征（JCIM,2019）。因此，我们认为在LXR配体新骨架基础上，运用深度学习的注意力机制可以识别选择性配体的特异性，发现高选择性的LXRβ配体。为了验证该假说，本项目拟研究如下内容：（1）分别建立已知LXRβ与α配体库；（2）在此基础上建立注意力机制的深度学习模型，发现LXRβ选择性配体的结构特征；（3）运用所建立的深度学习模型，在LXR配体新骨架基础上设计、合成和筛选LXRβ选择性配体；(4) 评价所发现的LXRβ选择性激动剂的抗胶质母细胞瘤的活性。本项目的顺利实施不仅为LXRβ选择性配体设计找到新方法，也为LXRβ与α配体选择性机制的最终解析奠定新的基础。

在已建立的基于深度学习、迁移学习及注意力机制的药物设计方法的基础上，我们扩充了现有LXR激动剂数据库，收集并整理了LXRbeta亚型选择性激动剂425个，通过子结构生成算法对优势片段进行拆分提取，并结合知识分析等手段，最终获得36个优势片段，其中24个对应子结合口袋A，12对应子结合口袋B。通过SyntaLinker连接算法产生针对靶标的聚焦化合物库，获得了3000多个潜在的选择性激动剂分子。进一步进行了虚拟筛选，选择了50个分子进行生物学验证实验，发现了新的具有抗胶质瘤活性的LXRβ选择性激动剂.