Bridging Genetically-Encoded Chemistry with Machine Learning to discover the properties of new Chemical matter

在这个项目中，我们将使用化合物的遗传编码库和这些化合物的空间表示的遗传编码来建立新的平台技术，允许预测一大类化合物和化学反应的化学、物理和生物性质。在过去的10年里，德尔达实验室开发了旨在加速分子发现的技术。我们开发了从展示在噬菌体表面的多肽中收集化学化合物的百万到十亿规模(106-109)的化学合成策略。方便的是，合成化合物的结构编码在噬菌体基因组的DNA序列中。因此，可以使用DNA测序来跟踪、选择和管理它们。我们的团队重新调整了下一代测序(NGS)-用于基因组测序和临床诊断的技术-允许对此类混合物中105-107个分子的结构和浓度进行常规测量。在使用我们的第三轮NSERC DG资金(2022-27年)后，我们将推进GE文库的应用，并在GE发现超慢反应(目标1)、GE糖聚糖库和镜像世界糖生物学中的机器学习(目标2)以及GE多价支架研究动态冰核过程(目标3)方面开发全新的方向。具体的研究计划将解决三个领域的基本问题：目标1探索发现和发展水中化学反应的新方法。通过将检测新反应的检测下限降低百万到十亿倍，我们将能够检测超慢反应，然后使用GE技术对其进行优化。我们预计，从不明显的超慢转化开始的方法将产生化学直觉可能不明显的反应。目标2使用遗传编码筛选和机器学习的组合来研究蛋白质和手性结构之间的分子识别。我们最近的兴趣是使用来自基因编码屏幕的现成和丰富的数据来训练ML模型。然而，这些数据的质量相当低。我们将使用与N个分子接触的交叉手性数据集和这些分子的N个镜像(对映体)来测试是否可以改善ML模型的质量。如果成功，这种方法可以极大地提高许多ML算法的训练质量，这些算法旨在解决涉及手性分子(分子识别、药物发现、催化)的问题。目标3继续对德尔达实验室于2015年启动的自催化冰核进行基础研究。我们将使用由M13噬菌体构建的遗传编码的精确多价结构来研究冰结合基序的空间呈现对冰核活动的作用。利用这一知识，我们将构建光触发的多价结构，当光照射过冷水时，这些结构可以使冰成核。

In this program, we will employ genetically-encoded libraries of chemical compounds and genetic-encoding of the spatial presentation of these compounds to build new platform technologies that allow predicting chemical, physical and biological properties of a wide class of chemical compounds and chemical reactions. In the last 10 years, the Derda Lab built technologies that aim to accelerate molecular discovery. We developed strategies for chemical synthesis of million-to-billion scale (106-109) collection of chemical compounds from peptides displayed on the surface of bacteriophage. Conveniently, structures of the synthesized compounds are encoded in DNA sequences in phage genome. Thus, they can be tracked, selected, and managed using DNA sequencing. Our group repurposed next-generation sequencing (NGS)-technology used for genome sequencing and clinical diagnostics-to allow for routine measurement of the structure and concentration of 105-107 molecules in such mixtures. Using our third cycle of NSERC DG funding (2022-27), we will advance the utility of GE-libraries and develop brand-new directions in GE discovery of ultra-slow reactions (Aim 1); GE glycan library and machine learning in mirror-world glycobiology (Aim 2) and GE multivalent scaffolds to study dynamic ice-nucleation processes (Aim 3). The specific research program will be solving fundamental problems in three areas: Aim 1 investigates new approach to discovery and development of chemical reactions in water. By lowering the detection limit for detection of new reactions by a factor of million to a billion, we will be able to detect ultra-slow reactions and then use GE-technology to optimize them. We anticipate that approaches that start from non-obvious ultra-slow transformations will give rise to reactions that might not be obvious to chemical intuition. Aim 2 investigates molecular recognition between proteins and chiral structures using a combination of genetically-encoded screening and machine learning. Our recent interest is in training ML models using readily available and abundant data from genetically-encoded screens. Quality of such data, however, is quite low. We will be testing whether the quality of ML models can be improved using cross-chiral datasets that contact N molecules and N mirror images (enantiomers) of these molecules. If successful, such approach can dramatically improve the quality of training of many ML algorithms that aim to solve problems that involve chiral molecules (molecular recognition, drug discovery, catalysis). Aim 3 continues fundamental investigation of the autocatalytic ice nucleation started by Derda Lab in 2015. We will use genetically-encoded precise multivalent architectures built from M13 phage to investigate the role of spatial presentation of ice-binding motifs on ice nucleation activity. Using this knowledge we will build the light-triggered multivalent constructs that nucleate ice in supercooled water upon irradiation with light.