耐药性感染每年导致超过100万人死亡。但大型制药公司基本上已经放弃了寻找新抗生素的努力，因为发现新抗生素所需的研发工作既耗时又昂贵。
Drug-resistant infections kill more than a million people each year. But big pharmaceutical companies have largely abandoned the quest for new antibiotics, because the research and development needed to discover them is time-consuming and expensive.
加拿大麦克马斯特大学斯托克斯实验室的博士生奥秋恩·阿诺德想出了一个解决方案：一个能预测哪些化合物可能会杀死危险细菌的人工智能模型。
Autumn Arnold, a PhD student in the Stokes Lab at McMaster University in Canada, came up with a fix: an AI model that predicts which chemical compounds might kill dangerous bacteria.
这个名为ESKAPE的模型是免费使用的：研究人员上传一份化合物名单，就能获得关于其抗菌活性的预测。“你只需要一个互联网连接，”运营该实验室的微生物学家乔纳森·斯托克斯说。研究人员保留对其发现的所有权，并且不必分享他们提交的化合物。
The model, called ESKAPE, is free to use: Researchers upload a list of compounds and get a prediction of their antibacterial activity. “All you need is an internet connection,” said Jonathan Stokes, a microbiologist who runs the lab. Researchers retain ownership of their discoveries, and don’t have to share the compounds they submit.
“我们面临着耐药性上升、抗生素研发管线严重枯竭以及巨大经济阻力的完美风暴，”斯托克斯说。“制造一种新药要花10亿美元，而且你永远无法收回成本。”到目前为止，研究人员已经使用ESKAPE模型筛查了超过1200万个分子。总体而言，该模型每天可以筛查超过4万种化合物，这有可能极大地加速药物发现。
“We have this perfect storm with rising resistance, a really lean antibiotic pipeline, and enormous economic headwinds,” Stokes said. “It’s going to cost you $1 billion to make a new drug, and you’re never going to make your money back.”So far, researchers have screened more than 12 million molecules using the ESKAPE model. In total, the model can screen more than 40,000 compounds a day, which has the potential to dramatically speed up drug discovery.
另一种由麻省理工学院和瑞典卡罗琳斯卡医学院的研究人员开发的方法，则是利用“生成式深度学习”来创造全新分子，以靶向抗生素耐药性。
Another approach, developed by researchers at MIT and Sweden’s Karolinska Institutet, is using “generative deep learning” to create entirely new molecules to target antibiotic resistance.
“这是一个全球性问题，”斯托克斯说。“我们需要每个人都来帮忙。”——佐伊·希弗
“This is a global problem,” Stokes says. “We need everybody helping out.”— Zoë Schiffer