Designing selective kinase inhibitors via deep learning

项目总结/摘要 现代癌症生物学在很大程度上依赖于激酶抑制剂，作为探测 灭活特定的激酶，但大多数常用的化学探针并不足以靶向- 选择性地对观察到的表型进行稳健的解释。通过组装大量的激酶 （对应于大部分人激酶组），已经可以确定对给定激酶组的选择性。 探针：然而，这些实验是昂贵的，并且在高通量下进行是不切实际的。我们有 最近开发了一种新的计算方法，用于快速准确地建立三维结构模型， 单个抑制剂/激酶复合物。该项目的目标1需要应用深度学习来构建模型， 预测单个抑制剂/激酶对的结合亲和力，使用来自以下的3D结构描述符 相应的抑制剂/激酶复合物。这个项目的目标2将建立非常大的计算库 新的化学物质，富含具有3D特性的化合物，补充激酶结合位点。我们 将首先使用目标1中开发的工具重新评估广泛使用的化学探针的选择性 细胞生物学家，从而告知哪些是有用的工具，哪些应该被弃用。到 为过时的化学探针提供替代品，我们将通过计算筛选Aim 2的库 对于更有选择性的化合物，首先关注CDK激酶和代表治疗作用的几种激酶， GIST中的漏洞。我们将综合得分最高的计算命中，并使用 生物化学测定、蛋白质组激酶组分析、结构生物学和细胞测定的升级。如果 如果成功，该项目将为几种迄今尚未解决的（“孤儿”）激酶提供新的化学探针， 作为化学工具和药物开发的起点。同样重要的是， 该项目将为设计有效和选择性激酶抑制剂提供一种可靠和有效的方法， 随后可用于开发针对500种人类激酶中每一种的新的高质量探针。

PROJECT SUMMARY/ABSTRACT Modern cancer biology leans heavily on kinase inhibitors as a means to probe the consequences of deactivating a particular kinase, but the majority of commonly-used chemical probes are not sufficiently target- selective for robust interpretation of the observed phenotypes. By assembling large panels of kinases (corresponding to much of the human kinome), it has become possible to determine the selectivity for a given probe: however, these experiments are expensive and impractical to perform at high throughput. We have recently developed a new computational approach for rapidly and accurately building 3D structural models of individual inhibitor/kinase complexes. Aim 1 of this project entails applying deep learning to build models for predicting the binding affinity of individual inhibitor/kinase pairs, using 3D structural descriptors derived from the corresponding inhibitor/kinase complexes. Aim 2 of this project will build very large computational libraries of novel chemical matter, enriched in compounds with 3D properties that complement kinase binding sites. We will first use the tools developed in Aim 1 to re-evaluate the selectivity of chemical probes that are widely used by cell biologists, thus informing on which ones are useful tools and which ones should be deprecated. To provide a replacement for the outdated chemical probes, we will computationally screen the libraries of Aim 2 for more selective compounds, focusing first on CDK kinases and several kinases that represent therapeutic vulnerabilities in GIST. We will synthesize the top-scoring computational hits, and characterize them using an escalation of biochemical assays, proteomic kinome profiling, structural biology, and cellular assays. If successful, this project will deliver new chemical probes for several hitherto unaddressed (“orphan”) kinases, to serve as chemical tools and as starting point for drug development. Equally importantly though, completion of this project will provide a robust and validated approach for designing potent and selective kinase inhibitors, to be subsequently applied for developing new high-quality probes against each of the 500 human kinases.