内容来源：https://www.wired.com/story/scientists-used-ai-to-create-16-new-viruses/

内容总结：

人工智能首次设计出全新病毒，或为对抗细菌耐药性开辟新路径

近日，一项发表在《科学》杂志上的研究显示，科学家首次利用人工智能系统设计出了一系列此前未知的病毒，这些病毒能够感染并消灭特定类型的细菌。这一突破为对抗细菌耐药性带来了新的可能性，但同时也引发了对该技术被滥用、用于设计生物武器的担忧。

多年来，科学家已经能够从头合成病毒，这些病毒通常用于研发和评估抗病毒药物及疫苗，并加深对微生物行为的理解。然而，此前合成病毒基因组主要依赖于复制已知病原体或其变体。

与此不同的是，斯坦福大学和Arc研究所的科学家开展的一项新研究，成功让人工智能基于自然界数百万种动物、植物、微生物、细菌和病毒的基因序列信息，设计出了简单、功能完备且前所未见的病毒。

研究人员以噬菌体为研究对象，这类微生物基因组相对较小，在受控条件下易于合成和操作。噬菌体仅感染细菌，因此具有巨大的生物技术应用潜力，有望成为对抗耐药菌感染、替代抗生素的有力工具。

这些全新病毒的创造基于Evo 1和Evo 2——专为计算生物学应用而设计的基座人工智能模型。这两个算法接受了来自生命各领域数百万个基因组的训练，目标是识别和学习复杂的进化模式，包括基因的典型组织方式、哪些序列是保守的，以及使生物体维持功能所需的生物学约束。

实验以能够感染大肠杆菌的噬菌体Phi X-174为参照。研究的目标并非复制该病毒，而是仅将其作为算法生成数千个全新基因组的指引，使这些基因组的遗传架构能够兼容感染大肠杆菌。

换言之，人工智能生成的病毒基因组保留了识别细菌、注入DNA、复制、产生新病毒颗粒并正确组装所必需的功能组织。然而，其具体DNA序列与自然界中存在的噬菌体存在显著差异。

16种新病毒问世

随后，科学家对人工智能生成的基因组进行了评估，筛选出最可能具有功能性的候选者，考虑因素包括基因组织、调控元件的存在以及其他参照Phi X-174噬菌体生物学特性的标准。

筛选后获得300个基因组样本，并在实验室中逐分子人工合成，随后导入大肠杆菌中，以检验它们能否产生功能性病毒。

在300个合成基因组中，仅有16个成功产生了功能完备的噬菌体，这些噬菌体拥有从未公开的序列、不同的基因、全新的调控元件，甚至基因组大小也有所不同。这些病毒的行为也各异：有的感染细菌更快，有的则表现出不同的复制能力。

这项本周发表于《科学》杂志的研究还评估了人工智能生成噬菌体对抗耐药菌的能力。实验将人工智能设计的噬菌体混合物与类似Phi X-174的天然噬菌体混合物，分别暴露于已对该病毒产生耐药性的大肠杆菌菌株中。

结果显示，人工智能生成的病毒能够迅速突破细菌耐药性并建立感染。作者表示，这一发现证明了“人工智能生成噬菌体疗法对抗快速进化细菌病原体的可行路径”。

里程碑的双面性

这一发现为应对日益严重的细菌耐药性问题提供了新的可能。研究人员指出，该方法有望推动个性化治疗的发展，使治疗方案能够以近乎与病原体相同的速度进化。

尽管这一里程碑标志着分子生物医学的重大进展，但也引发了对该技术被恶意利用的担忧，例如用于制造新型疾病、高毒性物质或可能引发新一轮大流行的病原体。

约翰斯·霍普金斯大学健康安全中心研究员莫里茨·汉克认为，目前尚缺乏能够有效阻止借助人工智能制造致命病毒的安全保障措施。他对《纽约时报》表示，科技进步的速度与有效监管框架的制定之间存在着“巨大的脱节”。

围绕这些风险的争论并不新鲜。三年前，兰德公司的一项研究就曾警告，当时最先进的人工智能系统已具备完善生物武器攻击策划与执行的能力。如今，随着该技术的快速发展，人们愈发担忧这类能力将变得更加强大和精密。

该非营利组织还警告称，人工智能系统演变的速度往往超过政府的监管能力。

本文最初发表于WIRED en Español，并经西班牙语翻译。

中文翻译：

人工智能系统首次创造出一系列此前未知、能够感染并消灭某些类型细菌的病毒。这一突破为对抗细菌耐药性开辟了新的可能，但也引发了对该技术被滥用、用于设计生物武器的担忧。

多年来，科学家已经能够从零开始合成病毒；这些病毒通常用于研发和评估抗病毒药物与疫苗，以及加深我们对这些微生物行为方式的理解。然而，这些病毒基因组的生成主要依赖于复制已知病原体或其变体。

相比之下，斯坦福大学和阿克研究所的科学家开展的一项新研究，成功让一个人工智能基于自然界中数百万动物、植物、微生物、细菌和病毒的基因序列所含信息，设计出了简单、功能完整且前所未见的病毒。

研究人员以噬菌体为研究对象——这类微生物基因组相对较小，在受控条件下较容易合成和操作。这些病毒只感染细菌，使其成为具有巨大生物技术潜力的工具，也是对抗耐药菌感染的一种有前景的抗生素替代方案。

这些全新病毒的创造基于Evo 1和Evo 2——为计算生物学应用而开发的基础人工智能模型。这两个算法经过数百万个来自生命各域基因组的训练，目标是识别和学习复杂的进化模式，包括基因通常如何组织、哪些序列是保守的，以及使生物体保持功能性的生物学约束条件。

实验设计以噬菌体Phi X-174（能够感染大肠杆菌）为参照。目标不是复制这种病毒，而是仅将其作为算法的指导，使其生成数千个全新的基因组，其遗传架构与感染大肠杆菌的能力兼容。

换言之，由人工智能生成的基因组衍生出的病毒，保留了识别细菌、注入其DNA、复制DNA、产生新病毒颗粒并正确组装所必需的功能组织。然而，具体DNA序列与自然界中观察到的噬菌体存在显著差异。

用AI创建了16种新病毒

随后，科学家评估了AI生成的基因组，以筛选出最可能具有功能性的候选，考虑因素包括基因组织、调控元件的存在，以及其他受Phi X-174噬菌体生物学启发而设定的标准。

经过筛选，最终得到300个基因组样本，并在实验室中逐分子人工合成。随后将它们导入大肠杆菌中，以测试它们是否能够产生功能性病毒。

在300个合成基因组中，只有16个产生了功能完整的噬菌体，其序列前所未有，基因不同，调控元件新颖，甚至基因组大小也各不相同。这些病毒的行为也各异：有些感染细菌更快，另一些则表现出不同的复制能力。

这项本周发表在《科学》杂志上的研究，还评估了AI生成的噬菌体对抗耐药细菌的能力。实验中，研究人员将一组AI设计的噬菌体和一组与Phi X-174相似的自然噬菌体，分别暴露于已经对该病毒产生耐药性的大肠杆菌菌株。

结果显示，AI生成的病毒能够迅速突破细菌耐药性并建立感染。据作者称，这一发现证明了“一条通往人工智能生成噬菌体疗法的路径，以应对快速演化的细菌病原体”。

这一里程碑的两面性

这一发现为解决日益严重的细菌耐药性问题开辟了新的可能。据研究人员称，这种方法有助于开发个性化治疗方案，使其能够以几乎与病原体本身相同的速度进化。

尽管这一里程碑代表了分子生物医学的重大进步，但它也引发了对该技术被恶意利用的担忧，例如用于开发新疾病、高毒性物质，或能够引发新疫情的病原体。

约翰斯·霍普金斯卫生安全中心研究员莫里茨·汉克认为，目前没有任何防护措施能够有效阻止借助AI制造致命病毒，他告诉《纽约时报》，科技进步的速度与有效监管框架的制定之间存在“巨大的脱节”。

围绕这些风险的争论并不新鲜。三年前，兰德公司的一项研究就曾警告，当时最先进的人工智能系统有能力完善生物武器攻击的策划与执行。如今，随着该技术的快速发展，人们越来越担心这些能力将变得更加强大和精密。

这家非营利组织还警告说，人工智能系统演进的速度往往超过政府的监管能力。

本文最初发表于《连线》西班牙语版，并已从西班牙语翻译。

英文来源：

For the first time, an artificial intelligence system has created a series of previously unknown viruses capable of infecting and eliminating certain types of bacteria. This breakthrough opens up new possibilities for combating bacterial resistance. However, it also raises concerns about the potential misuse of this technology to design biological weapons.
For several years now, scientists have been able to synthesize viruses from scratch; these are typically used to develop and evaluate antiviral drugs and vaccines, as well as to expand our understanding of how these microorganisms behave. However, the production of these viral genomes has primarily relied on replicating previously known pathogens or variants.
In contrast, a new study conducted by scientists at Stanford University and the Arc Institute succeeded in having an AI design simple, functional, and previously unseen viruses based on information contained in the genetic sequences of millions of animals, plants, microbes, bacteria, and viruses found in nature
The researchers worked with bacteriophages—microorganisms characterized by relatively small genomes, which are relatively easy to synthesize and manipulate under controlled conditions. These viruses infect only bacteria, making them a tool with enormous biotechnological potential and a promising alternative to antibiotics for combating resistant bacterial infections.
The creation of these entirely new viruses was based on Evo 1 and Evo 2, foundational AI models developed for computational biology applications. Both algorithms were trained on millions of genomes from all domains of life, with the goal of identifying and learning complex evolutionary patterns, including how genes are typically organized, which sequences are conserved, and the biological constraints that allow an organism to remain functional.
The experimental design used the bacteriophage Phi X-174—which is capable of infecting the bacterium Escherichia coli (E. coli)—as a reference. The goal was not to reproduce this virus, but rather to use it solely as a guide for the algorithms to generate thousands of completely new genomes with a genetic architecture compatible with infecting E. coli.
In other words, the viruses derived from the genomes created by the AI retained the functional organization essential for recognizing the bacterium, inserting their DNA, replicating it, producing new viral particles, and assembling them correctly. However, the specific DNA sequences differed considerably from those observed in naturally occurring bacteriophages.
16 New Viruses Created Using AI
The scientists then evaluated the AI-generated genomes to select those most likely to be functional, taking into account factors such as gene organization, the presence of regulatory elements, and other criteria inspired by the biology of the Phi X-174 bacteriophage.
This selection resulted in a sample of 300 genomes, which were artificially synthesized, molecule by molecule, in the laboratory. They were then introduced into E. coli bacteria to test whether they were capable of producing functional viruses.
Of the 300 synthesized genomes, only 16 gave rise to fully functional bacteriophages, featuring previously unpublished sequences, different genes, new regulatory elements, and even varying genome sizes. The behavior of these viruses also varied: While some infected the bacteria more quickly, others exhibited different abilities to replicate.
The research, published this week in the journal Science, also evaluated the ability of AI-generated bacteriophages to combat resistant bacteria. The experiment involved exposing a mixture of AI-designed phages and a mixture of natural phages similar to Phi X-174 to strains of E. coli that had already developed resistance to that virus.
The results showed that the AI-generated viruses were able to rapidly overcome bacterial resistance and establish infection. According to the authors, this finding demonstrates “a path toward artificial intelligence–generated phage therapies against rapidly evolving bacterial pathogens.”
The Two Sides of the Milestone
The discovery opens up new possibilities for tackling the growing problem of bacterial resistance. According to the researchers, this approach could facilitate the development of personalized treatments capable of evolving at nearly the same rate as the pathogens themselves.
Although this milestone represents a significant advance for molecular biomedicine, it also raises concerns about the potential malicious use of this technology to develop, for example, new diseases, highly toxic substances, or pathogens capable of triggering a new pandemic.
Moritz Hanke, a researcher at the Johns Hopkins Center for Health Security, argues that there are currently no safeguards capable of effectively preventing the creation of a lethal virus with the help of AI, telling The New York Times that there is “a huge disconnect” between the speed at which science and technology are advancing and the development of effective regulatory frameworks.
The debate surrounding these risks is not new. Three years ago, a study by the Rand Corporation warned that the most advanced AI systems at the time had the capacity to refine the planning and execution of attacks using biological weapons. Now, with the rapid development of this technology, fears are growing that such capabilities will become even greater and more sophisticated.
The nonprofit organization also warned that the speed at which AI systems evolve often outpaces governments’ capacity for regulatory oversight.
This story originally appeared on WIRED en Español and has been translated from Spanish.
Comments
Back to top