Tackling the pandemic of antibiotic-resistant infections: An artificial intelligence approach to new druggable therapeutic targets and drug discovery

很难想象抗生素被发现之前的生活。肺结核、肺炎和百日咳等感染是常见的杀手，如果轻微的伤口和烧伤感染，它们是致命的。使用抗生素控制细菌感染也许是现代医学最重要的成就。然而，我们未能跟上微生物对现有治疗方法越来越耐药的步伐。2019冠状病毒病大流行表明，如果没有有效治疗，感染将对人类健康构成威胁。抗生素耐药感染已成为另一种全球流行病，每年在全球造成近500万人死亡。特别令人关切的是肺炎克雷伯菌引起的感染，在全球范围内，它是与任何抗生素耐药感染导致的死亡（25万人）相关的第三大病原体。越来越多的对“最后手段”抗微生物药物具有耐药性的菌株的隔离大大缩小了治疗选择，或在某些情况下完全消除了治疗选择。这在低收入和中等收入国家尤其令人担忧。不幸的是，没有发明新的药物类别，耐药性继续无情地蔓延。风险很高，我们可能正在进入前抗生素时代。据英国公共卫生部计算，由于缺乏有效的抗生素，300多万例手术和癌症治疗将危及生命。据估计，在未来30年里，英国将有9万多人死于抗生素耐药性感染。抗生素药物发现的黄金时代利用了土壤微生物产生的抗菌产品，但这种方法在经过20年的系统筛选后已经耗尽。研究人员挖掘了不同来源的天然产物，如海洋环境、植物，甚至是居住在我们肠道中的无害微生物群落，并取得了令人鼓舞的结果。然而，没有一种分离出来的化合物进入药物开发。更好地了解微生物抵抗抗生素的手段可能会导致发现迄今为止未知的目标，适合开发新药。在这项研究中，我们将使用人工智能从肺炎克雷伯菌中识别新的潜在药物靶点，当这些靶点被阻断时，可能会使微生物对抗生素敏感，甚至可能促进我们的防御系统对克雷伯菌的清除。我们将训练有监督的学习者，让他们通过我们在实验室生成的数据，读取微生物的基因组，找到研究人员忽视的目标。接下来，利用其他学习器，我们将确定可以阻断这些目标的药物。具体来说，我们将搜索已经批准用于人类但与抗菌活性无关的药物。我们将在实验室进行实验，以确认这些药物的效果。从药物发现的角度来看，我们的方法大大缩短了药物开发过程，从而允许从基础研究到临床开发的潜在快速过渡。我们设想我们的结果将鼓励其他学者和制药公司遵循这一新的研究途径，以解决缺乏对抗生素耐药微生物治疗的问题。为了促进这一点，我们将免费提供我们的协议、模型和数据。

It is difficult to imagine life before antibiotics were discovered. Infections such as tuberculosis, pneumonia and whooping cough were common killers - and if minor wounds and burns became infected they were fatal. The use of antibiotics to control bacterial infections is perhaps the most important achievement of modern medicine. However, we have failed to keep pace with microbes becoming increasingly resistant to available treatments. The Covid-19 pandemic exemplifies the threat to human health of an infection without an effective treatment. Antibiotic-resistant infections are already another global pandemic claiming almost 5 million deaths per year globally. Of particular concern are the infections caused by Klebsiella pneumoniae, globally, the third leading pathogen associated with deaths (250 000) attributed to any antibiotic-resistant infection. The increasing isolation of strains resistant to "last resort" antimicrobials has significantly narrowed, or in some settings completely removed, the therapeutic options. This is particularly alarming in low and middle-income countries. Unfortunately, new classes of drugs are not being invented and resistance continues to spread inexorably. The stakes are high and we might be entering into a pre-antibiotic era. Public Health England has calculated that the lack of effective antibiotics will render more than the three million operations and cancer treatments life-threatening, and more than 90,000 people are estimated to die in the UK over the next 30 years due to antibiotic-resistant infections.The golden era in antibiotic drug discovery leveraged the antibacterial products produced by soil microorganisms but this approach became exhausted after 20 years of systematic screening. Researchers have mined different sources of natural products such as marine environments, plants, and even the community of harmless microbes inhabiting our gut with encouraging results. Yet, none of the compounds isolated have entered into drug development. A better understanding of the means used by microbes to resist antibiotics may result in the discovery of hitherto unknown targets suitable to develop new drugs against. In this research, we will use artificial intelligence to identify new potential druggable targets from K. pneumoniae that when blocked may render the microbe susceptible to antibiotics and perhaps may even facilitate the clearance of Klebsiella by our defenses. We will train supervised learners to go through data we will generate in the laboratory and to read the genome of the microbe to find these targets that researchers have overlooked. Next, and utilizing other learners, we will identify drugs that can block these targets. Specifically, we will search drugs already approved for use in humans but used for purposes unrelated to antimicrobial activity. We will carry out experiments in the laboratory to confirm the effect of these drugs. From the drug discovery point of view, our approach significantly shortcuts the drug development process hence allowing a potential fast-track transition from the basic research to clinical development. We envision that our results will encourage other academics as well as pharmaceutical companies to follow this new avenue of research to tackle the problem of the lack of therapies for microbes resistant to antibiotics. To facilitate this, we will make freely available our protocols, models and data.