Deep learning based antibody design using high-throughput affinity testing of synthetic sequences

使用合成序列的高通量亲和力测试进行基于深度学习的抗体设计

基本信息

批准号：
10362725
负责人：
David K Gifford
金额：
$ 13.97万
依托单位：
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
依托单位国家：
美国
项目类别：
财政年份：
2018
资助国家：
美国
起止时间：
2018-03-09 至 2025-02-28
项目状态：
未结题

来源：
https://reporter.nih.gov/project-details/10362725
关键词：
Affinity Animals Antibodies Antibody Affinity Antigens Architecture Binding Biological Assay Budgets Classification Cloud Computing Communicable Diseases Computing Methodologies DNA Sequence Data Data Set Disease Goals Human Immunize Immunotherapeutic agent Learning Machine Learning Malignant Neoplasms Methodology Methods Modeling Molecular Machines Oligonucleotides Output Performance Phage Display Property Randomized Research Services Specific qualifier value Specificity Statistical Models Technology Test Result Testing Therapeutic Thinness Time Training Treatment Efficacy Update Virus Diseases Work cloud based commercialization computing resources cost deep learning design experimental study human disease improved iterative design learning strategy machine learning method machine learning model manufacture mathematical methods molecular dynamics novel novel strategies outcome prediction receptor

项目摘要

Project Summary We will develop and apply a new high-throughput methodology for rapidly designing and testing antibodies for a myriad of purposes, including cancer and infectious disease immunotherapeutics. We will improve upon current approaches for antibody design by providing time, cost, and humane benefits over immunized animal methods and greatly improving the power of present synthetic methods that use randomized designs. To accomplish this, we will display millions of computationally designed antibody sequences using recently available technology, test the displayed antibodies in a high-throughput format at low cost, and use the resulting test data to train molecular dynamics and machine learning methods to generate new sequences for testing. Based on our test data our computational method will identify sequences that have ideal properties for target binding and therapeutic efficacy. We will accomplish these goals with three specific aims. We will develop a new approach to integrated molecular dynamics and machine learning using control targets and known receptor sequences to refine our methods for receptor generalization and model updating from observed data (Aim 1). We will design an iterative framework intended to enable identification of highly effective antibodies within a minimal number of experiments, in which our methods automatically propose promising antibody sequences to profile in subsequent assays (Aim 2). We will employ rounds of automated synthetic design, affinity test, and model improvement to produce highly target-specific antibodies. (Aim 3). !

项目摘要我们将开发并应用一种新的高通量方法来快速为无数目的设计和测试抗体，包括癌症和传染病免疫治疗药。我们将改善当前通过提供时间，成本和人道益处来实现抗体设计的方法超越免疫动物方法，并大大提高了现在的力量使用随机设计的合成方法。为此，我们将显示数百万使用的计算设计抗体序列最近使用可用技术，以高通量格式测试显示的抗体低成本，并使用最终的测试数据训练分子动力学和机器学习方法生成用于测试的新序列。基于我们测试数据我们的计算方法将确定具有理想的序列目标结合和治疗功效的特性。我们将完成这些具有三个特定目标的目标。我们将开发一种综合的新方法使用控制目标和已知的分子动力学和机器学习受体序列以完善我们的受体概括和模型的方法从观察到的数据更新（AIM 1）。我们将设计一个迭代框架旨在鉴定最小的高效抗体我们的方法自动提出有希望的实验数量在随后的测定中对剖面的抗体序列（AIM 2）。我们将雇用自动合成设计，亲和力测试和模型改进产生高靶标特异性抗体。（目标3）。呢