Annotating dark ion-channel functions using evolutionary features, machine learning and knowledge graph mining

使用进化特征、机器学习和知识图挖掘注释暗离子通道函数

• 批准号: 10457684
• 负责人: Natarajan Kannan
• 金额: $ 47.79万
• 依托单位: UNIVERSITY OF GEORGIA
• 依托单位国家: 美国
• 财政年份: 2022
• 资助国家: 美国
• 起止时间: 2022-07-07 至 2024-06-30
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/10457684
• 关键词: Address; Amino Acid Sequence; Animal Model; Artificial Intelligence; Bayesian Method; Biological Process; Blindness; Calcium; Cardiovascular Diseases; Cell physiology; Chemicals; Classification; Communities; Cryoelectron Microscopy; Data; Disease; Electrophysiology (science); Epilepsy; Evolution; Family; Family member; Felis catus; Foundations; G-Protein-Coupled Receptors; Genes; Genome; Goals; Graph; Homeostasis; Human; Human Genome; Informatics; Ion Channel; Kidney Failure; Knowledge; Language; Light; Link; Machine Learning; Malignant Neoplasms; Maps; Methods; Mining; Mission; Molecular; Mutation; Mutation Analysis; Names; Organism; Orthologous Gene; Outcome; Pathway interactions; Pattern; Phosphotransferases; Physiological; Play; Protein Family; Proteins; Proteome; Readability; Regulation; Research; Research Personnel; Resources; Role; Semantics; Source; Structural Models; Structure; Structure-Activity Relationship; Systems Biology; Testing; Text; Training; base; cell type; deep learning; deep learning model; drug discovery; experimental study; genome resource; human disease; interest; knowledge graph; nervous system disorder; novel; patch clamp; protein protein interaction; structural biology; tool

Project Summary The overall goal of this proposal is to annotate understudied dark ion channels using a combination of computational and experimental approaches. Our working hypothesis is that the wealth of evolutionary data encoded in ion-channel sequences from diverse organisms and integrative mining of evolutionary data with structure, function, pathway and expression data will provide important context for predicting and annotating dark channel functions at the molecular and cellular level. As a preliminary test of our hypothesis, we have generated a functional classification of ion channel sequences using a protein language based deep learning model trained on 250 million protein sequences and have delineated the distinguishing sequence and structural features of understudied Calcium Homeostasis Modulator (CALHM) family. We have also built an integrated Knowledge Graph (KG) linking diverse forms of ion channel information in machine readable format and deployed the KG for predicting physiological functions using a graph embedding approach that efficiently captures contextual information encoded in large graphs. We propose to build on these successful studies to accomplish the following two aims. Aim1 will develop new tools and resources for visualizing, mining and annotating dark channels using evolutionary features and structural models made available through cryo-EM studies and artificial intelligence based structure prediction methods. The unique modes of CALHM family gating and oligomerization mechanisms predicted through evolutionary studies will be experimentally validated through mutational studies and electrophysiology experiments. Aim2 will further develop the ion channel KG by semantically linking multiple disparate sources of data including cell-type specific expression, orthologs from model organisms and electrophysiology parameters. Knowledge graph embedding approaches will be employed to predict links between understudied channels, disease associations and physiological functions and the predictions will be made available as text summaries in the IDG resource Pharos. The proposed studies are expected to address the unique informatics needs of the ion channel community by providing new tools and resources for mapping sequence-structure-function relationships. The proposed studies will also provide new testable hypotheses on understudied channels and significantly enhance the value of Pharos in illuminating the functions of the understudied druggable proteome.

项目摘要 这项提议的总体目标是使用计算和实验相结合的方法来注释未被研究的暗离子通道。我们的工作假设是，从不同生物的离子通道序列中编码的丰富的进化数据，以及对进化数据与结构、功能、途径和表达数据的综合挖掘，将为在分子和细胞水平上预测和诠释暗通道功能提供重要的背景。作为对我们假设的初步检验，我们使用基于蛋白质语言的深度学习模型对离子通道序列进行了功能分类，该模型对2.5亿个蛋白质序列进行了训练，并描绘了未被研究的钙稳态调节器(CALHM)家族的区别序列和结构特征。我们还构建了一个集成的知识图(KG)，以机器可读的格式连接不同形式的离子通道信息，并使用图形嵌入方法部署用于预测生理功能的KG，该方法可以有效地捕获编码在大图中的上下文信息。我们建议在这些成功研究的基础上，实现以下两个目标。AIM1将开发新的工具和资源，用于可视化、挖掘和注释暗通道，使用通过低温电磁研究和基于人工智能的结构预测方法获得的进化特征和结构模型。通过进化研究预测的CALHM家族门控和寡聚机制的独特模式将通过突变研究和电生理学实验进行实验验证。AIM2将进一步开发离子通道KG，通过语义链接多个不同的数据来源，包括细胞类型特定表达、模式生物的同源基因和电生理参数。知识图谱嵌入方法将被用来预测未被研究的渠道、疾病关联和生理功能之间的联系，预测将作为文本摘要在IDG资源Pharos中提供。拟议的研究有望通过提供绘制序列-结构-功能关系的新工具和资源来满足离子通道社区的独特信息学需求。拟议的研究还将为未被研究的通道提供新的可验证的假说，并显著提高PHAROS在阐明未被研究的可药物蛋白质组的功能方面的价值。