While working on a COVID-19-related project during the lockdown, Kärt Tomberg, PhD, found herself thinking about introns. She was part of a team working on the spike protein used in vaccines. Her task was to make the expression construct for the protein. “I’m a geneticist and I’d been looking at genes for 19 years but I’d never had to design the transgene,” she told .

在封锁期间，Kärt Tomberg博士在从事与新型冠状病毒相关的项目时，发现自己在思考内含子。她是研究疫苗中使用的刺突蛋白团队的一员。她的任务是为蛋白质构建表达构建体。她说：“我是一名遗传学家，研究基因已经19年了，但我从未设计过转基因。”。

. Searching through S-protein designs shared by other scientists, she could not quite figure out why they were designed the way they were.

。通过搜索其他科学家共享的S蛋白设计，她不太明白为什么它们会被设计成这样。

“Fifty years ago, [before] we sequenced human genomes let alone other genomes, we had no idea how genes looked,” she said. Yet “we are using genes and gene structures that look more like viruses and bacterial structure than human genes. Surely we could make them look more like human genes.” A clear difference is that natural viral genes do not have introns.

她说：“五十年前，在我们对人类基因组进行测序之前，更不用说其他基因组了，我们不知道基因是什么样子的。”。然而，“我们使用的基因和基因结构比人类基因更像病毒和细菌结构。当然，我们可以使它们看起来更像人类基因。”一个明显的区别是，天然病毒基因没有内含子。

A self-described “classic nerd,” Tomberg began running some experiments of her own to try “to figure out what makes an intron from the spliceosomes point of view” and “to predict it,” something which really had not been done before, she said. .

作为一个自称“经典书呆子”的汤姆伯格，她开始自己进行一些实验，试图“从剪接体的角度找出内含子的组成”并“预测它”，这是以前从未做过的事情，她说。。

It took several months but she eventually came up with a way to boost protein expression from viral genes by inserting introns. When she tested her approach on spike protein expression, there was an almost 10-fold increase in protein production. Her work forms the basis of the technology that underlies ExpressionEdits, a company co-founded by Tomberg and genome engineering expert Allan Bradley, PhD.

花了几个月的时间，但她最终找到了一种通过插入内含子来提高病毒基因蛋白质表达的方法。当她测试她的刺突蛋白表达方法时，蛋白质产量增加了近10倍。她的工作构成了ExpressionEdits技术的基础，ExpressionEdits是一家由汤姆伯格和基因组工程专家艾伦·布拉德利博士共同创立的公司。

The company’s Genetic Syntax Engine is an artificial intelligence-powered platform that uses “intronization” to enhance gene expression. Specifically, it creates in-cDNA with optimal sites and introns that increase protein production without altering the underlying genetic sequence..

该公司的遗传语法引擎是一个人工智能平台，它使用“内含”来增强基因表达。具体而言，它在具有最佳位点和内含子的cDNA中产生，从而在不改变潜在遗传序列的情况下增加蛋白质产量。。

This week, the company announced its first collaboration and licensing agreement with a major pharmaceutical company, Boehringer Ingelheim, that is focused on developing enhanced gene therapies. Under the terms of the partnership, Boehringer will leverage ExpressionEdits’ Genetic Syntax Engine, an AI-powered platform that generates optimized introns that boost protein expression without altering the underlying genetic sequence.

本周，该公司宣布与一家大型制药公司勃林格殷格翰（BoehringerIngelheim）签订了第一份合作和许可协议，该协议专注于开发增强型基因疗法。根据合作条款，勃林格将利用ExpressionEdits的遗传语法引擎，这是一个人工智能驱动的平台，可以生成优化的内含子，从而在不改变潜在遗传序列的情况下促进蛋白质表达。

The partners are not disclosing which therapies they will collaborate on at this time. .

合作伙伴目前没有透露他们将合作的疗法。。

The inner workings of ExpressionEdits’ engine are complex but Tomberg offered some insights into how the platform makes predictions. “Nature doesn’t have perfect introns [meaning] it doesn’t care to splice carefully or to express things in a single form. If I trained a model to predict a good intron [using] data that is already in nature, we wouldn’t get good rules,” she explained.

ExpressionEdit引擎的内部工作很复杂，但Tomberg提供了一些关于平台如何进行预测的见解。她解释说：“自然界没有完美的内含子（意思是说），它不在乎仔细拼接或以单一形式表达事物。如果我训练一个模型来预测一个好的内含子（使用）自然界已有的数据，我们就不会得到好的规则。”。

That meant that “we had to make the data ourselves.”  .

这意味着“我们必须自己制作数据”。

Scientists at ExpressionEdits spent the last two years building a training dataset by generating billions of data points on different combinations of introns and exons that could be used to learn patterns that indicate good splicing and bad splicing. They then used this dataset to train AI models to make predictions about where introns should be added, how many are needed, which particular introns work best, and what should happen to the surrounding exons.

ExpressionEdits的科学家在过去两年中通过在内含子和外显子的不同组合上生成数十亿个数据点来构建训练数据集，这些数据点可用于学习指示良好剪接和不良剪接的模式。。

Sometimes, the splicing decision makes sense because it removes something toxic, Tomberg said. In other cases, it’s less clear why a particular prediction results in greater protein production suggesting that the computer may be picking up a biological pattern that scientists don’t understand yet. .

汤姆伯格说，有时候，剪接决定是有道理的，因为它可以去除有毒物质。在其他情况下，不太清楚为什么一个特定的预测会导致更高的蛋白质产量，这表明计算机可能正在采集科学家尚不了解的生物模式。。

The default setting for the system is to generate a single intron design but there is room for customization. For example, it can be instructed to generate multiple designs but to restrict them to a specific number of base pairs. It can also generate designs based on datasets on codons found in nature rather than ExpressionEdits’ internally generated data, Tomberg said. .

系统的默认设置是生成单个内含子设计，但有定制的空间。例如，可以指示生成多个设计，但将其限制为特定数量的碱基对。汤姆伯格说，它还可以根据自然界中发现的密码子数据集而不是ExpressionEdit内部生成的数据来生成设计。。

Besides the Boehringer deal, ExpressionEdits is open to exploring other relationships with others in the gene therapy space, specifically larger, more mature companies. “We don’t have the capacity to develop the drugs,” Tomberg said. “We want to partner with someone who has the capability of taking these candidates all the way because then we can learn the most about our technology within that space.” Also given the current uncertainty in the gene therapy market, for a small company of about 14 people, “we have to be focused on who we choose.” .

除了勃林格协议之外，ExpressionEdit还愿意探索与基因治疗领域其他公司的其他关系，特别是更大、更成熟的公司。汤姆伯格说：“我们没有能力开发这些药物。”。“我们想与有能力全程录取这些候选人的人合作，因为这样我们就可以在该领域内对我们的技术了解得最多。”同时考虑到目前基因治疗市场的不确定性，对于一家大约14人的小公司来说，“我们必须专注于我们选择的人。”。