Peptide biosynthesis off the beaten path: Machine learning-guided discovery of non-canonical peptide natural products

超过50%的药物是天然产物，或者至少是受天然产物的启发。后基因组时代，对天然产物生物合成的深入了解和可用基因组序列信息数量的不断增加，导致基因组挖掘作为一门新的学科被引入，用于天然产物的靶向鉴定。基因组挖掘是一种利用基因组序列信息来评估生物体天然产物生物合成潜力的计算机天然产物发现策略。一些高度复杂的基因组挖掘平台已经开发出来，用于鉴定和注释微生物基因组序列中的教科书天然产物生物合成基因簇（BGCs）。目前的基因组挖掘管道无法识别相应的bgc的天然产物，其生物合成并不严格遵循为每种天然产物类别建立的看似普遍的生物合成原则，这表明，目前可用的生物信息学算法无法识别一定比例的非规范bgc。这些非规范的bgc显示了一个几乎未开发的宝藏，用于鉴定真正新颖的天然产物支架和前所未有的生化转化。我的团队将开发基于机器学习的基因组挖掘算法，用于有针对性地识别这些非规范的bgc。我们将筛选所有公开可用的基因组序列，以寻找(1)迄今为止被忽视的核糖体合成和翻译后修饰肽（RiPP） BGCs家族，(2)前所未有的非核糖体肽合成酶（NRPS）和多酮合成酶BGCs，它们具有隐酶结构域或前所未有的模块结构，以及(3)编码以RiPP和NRPS独立方式生物合成肽的酶的BGCs。将对选定的BGCs进行重构，并将每个基因置于不同的小分子诱导启动子的控制下。重组后的bgc随后将在高度优化的异源宿主生物中表达。天然产物将被纯化，纯化后的代谢产物的结构将被阐明。生物合成研究将通过一次抑制一个基因的转录来进行，并将研究对产物形成的影响。这种不依赖基因敲除的方法使我们能够提出生物合成模型，并产生天然产物中间体、类似物和分流产物的小文库。这些小的天然产物库将接受广泛的生物活性分析，以确定潜在的药物相关性的天然产物，并进行初步的SAR研究。该研究有助于绘制天然产物生物合成暗物质图谱，拓展肽天然产物的化学空间，并导致前所未有的生化转化的鉴定和表征。

More than 50% of all drugs are natural products or have at least been inspired by natural products. Insights into the biosynthesis of natural products and the ever-increasing number of available genome sequence information in the post genomics era have resulted in the introduction of genome mining as a new discipline for the targeted identification of natural products. Genome Mining is an in-silico natural product discovery strategy that uses genome sequence information to assess the natural product biosynthetic potential of an organism. Several highly sophisticated genome mining platforms have been developed for the identification and annotation of textbook natural product biosynthetic gene clusters (BGCs) in microbial genome sequences. The discovery of natural products for which the corresponding BGCs cannot be identified by current genome mining pipelines and whose biosynthesis does not strictly follow the seemingly universal biosynthetic principles established for every natural product class, suggests that a proportion of non-canonical BGCs escapes unrecognized by currently available bioinformatic algorithms. These non-canonical BGCs display an almost untapped treasure trove for the identification of truly novel natural product scaffolds and unprecedented biochemical transformations. My group will develop machine learning-based genome mining algorithms for the targeted identification of these non-canonical BGCs. We will screen all publicly available genome sequences for the presence of (1) so far overlooked families of ribosomally synthesized and posttranslationally modified peptide (RiPP) BGCs, (2) unprecedented non-ribosomal peptide synthetase (NRPS) and polyketide synthase BGCs that harbor cryptic enzymatic domains or unprecedented module architectures and (3) BGCs which encode enzymes that biosynthesize peptides in a RiPP and NRPS-independent manner. A selection of the identified BGCs will be refactored and each gene placed under the control of a different, small molecule inducible promoter. The refactored BGCs will subsequently be expressed in highly optimized heterologous host organisms. Natural products will be purified and the structures of the purified metabolites elucidated. Biosynthetic studies will be conducted by repressing the transcription of one gene at the time and the effect on product formation will be studied. This gene knockout-independent approach allows us to propose biosynthetic models and yields small libraries of natural product intermediates, analogs and shunt products. These small natural product libraries will be subjected to a broad panel of bioactivity assays to identify natural products of potential pharmaceutical relevance and to conduct initial SAR studies. This research helps chart natural product biosynthetic dark matter, expands the chemical space of peptide natural products and leads to the identification and characterization of unprecedented biochemical transformations.