Application of Artificial Intelligence-Driven Design of Function-Directed Ligands for Selective Retinoic Acid Receptor Binding

视黄酸（RA）配体与视黄酸受体（RAR）类核受体结合。RAR配体的形状和结构，虽然最佳激活RAR的是知之甚少，这个项目的目的是建模RAR到目前为止还没有达到能够设计激活受体的配体，并了解基因组和非基因组信号的RAR的触发途径。为此，将开发新的方法，这些方法不仅适用于RAR，而且适用于核受体类别的许多成员。我们是了解RA在大脑中功能的世界领导者（1），发现RAR配体作用的几种机制对其作用至关重要：基因组活性，打开基因转录，和快速的非基因组作用，涉及激酶激活。在这个项目中，一个完全不同的方法将被用于配体设计，建模和理解不同RAR的结合选择性。这将结合使用分子对接、原子分子动力学模拟和机器学习技术，超越静态的2D或3D配体描述符，开发复杂的定量构效关系（QSAR）模型，该模型将动力学与形状和化学选择性结合起来。所采用的技术将包括配体设计的人工智能方法，包括使用结构域-具体的技术，如DeepChem和更通用的工具，如Keras和TensorFlow。从化学和生物学方面，将制备从上述工作预测的合成类视色素，并将其应用于RAR活性的各种测定中，例如转录活性、通过各种激酶的非基因组信号传导以及使用细胞系和原代神经细胞的蛋白质翻译控制。对RAR的配体结合口袋以及如何设计不同的配体以触发不同的分子途径的完整理解可能具有未来设计用于治疗神经退行性疾病的配体的潜力。该项目是高度协作和跨学科的，涉及不同大学的研究人员的大型和多样化的联盟，以及一些工业合作伙伴，允许核受体基础科学的所有分支得到解决。该学生将致力于采用人工智能和建模，与伦敦大学学院的Coveney合作设计新的RAR激活剂配体，与达勒姆的Whiting进行药物靶向合成，并与阿伯丁的McCaffery和Greig进行生物学研究。通过这个项目，学生将熟悉基于人工智能的技术（如卷积神经网络，随机森林和支持向量机）来研究和操纵受体蛋白，并更全面地了解RAR的配体激活和核受体功能。参考文献：1.希勒KD，斯托尼PN，摩根PJ，麦卡弗里PJ。对大脑有益的维生素。神经科学趋势2012;35：733-41

Retinoic acid (RA) ligands bind to the retinoic acid receptor (RAR) class of nuclear receptor. The shape and structure of RAR ligands though that optimally activate RARs is poorly understood and this project aims to model RARs to a degree not so far obtained to be able to design ligands that activate the receptors and understand the triggering routes for RARs for both genomic and non-genomic signalling. New approaches will be developed to these ends that will be applicable not just to RARs but to many members of the nuclear receptor class of receptors. We are world-leaders in understanding the function of RA in the brain (1), making the discovery that several mechanisms by which RAR ligands act are crucial for their action: both genomic activity, to turn on gene transcription, and rapid non-genomic action, involving kinase activation.In this project a radically different approach will be taken to ligand design, to modelling and to understanding binding selectivities to the different RARs. This will use a combination of molecular docking, atomistic molecular dynamics simulations and machine learning techniques, to move beyond static 2D or 3D ligand descriptors and develop complex Quantitative Structure-Activity Relationship (QSAR) models which incorporate dynamics alongside shape and chemical selectivity.The techniques employed will include an AI approach to ligand design including the use of domain-specific technologies such as DeepChem and more generic tools such as Keras and TensorFlow. From the chemical and biological side, synthetic retinoids predicted from the above work will be prepared and applied in a variety of assays for RAR activity such as transcriptional activity, non-genomic signalling via a variety of kinases and control of protein translation using cell lines and primary neural cells. A complete understanding of the ligand binding pocket of RAR and how different ligands may be designed to trigger different molecular pathways may have future potential for design of ligands for the treatment of neurodegenerative diseases. The project is highly collaborative and interdisciplinary, involving a large and diverse consortium of researchers at different Universities, and a number of industrial partners, allowing all branches of the fundamental science of nuclear receptors to be addressed. The student will work on employing AI and modelling, designing new RAR activator ligands in collaboration with Coveney at UCL, drug target synthesis with Whiting at Durham, and biology with McCaffery and Greig at Aberdeen. The shape and properties of the designed drugs will be correlated with their biological function.From this project, the student will become familiar with AI based techniques (such as convolutional neural networks, random forest and support vector machines) to study and manipulate receptor proteins and understand more completely ligand activation of RAR and nuclear receptor function. References:1. Shearer KD, Stoney PN, Morgan PJ, McCaffery PJ. A vitamin for the brain. Trends Neurosci. 2012;35:733-41