'Resurrection' of ancient biological molecules, AI solves antibiotic resistance, two papers published by Fudan University and Penn University collaborative teams were published in Cell and Nature sub-journals

編輯| 蘿蔔皮

抗生素抗藥性感染每年在全球造成約127 萬人死亡，預計到2050 年，如果沒有特效的新藥，每年死亡人數將達到1000 萬人，因此需要採取緊急措施來應對抗生素抗藥性。

賓州大學的校長助理教授（Presidential Assistant Professor） Cesar de la Fuente 說：「即使感覺身體好些了，也要確保完成抗生素療程，這是許多人聽過，但經常忽視的醫學口頭禪。 「近幾十年來，這導致了抗藥性細菌的增加，全球健康危機日益嚴重，每年造成約495 萬人死亡，甚至可能使普通感染也致命。」

##De la Fuente 和復旦大學、賓州大學的研究人員組成的跨領域研究團隊，一直致力於研究應對抗生素抗藥性問題。

在最新的研究中，他們開發了一種人工智慧工具來挖掘龐大且基本上未開發的生物數據——超過1000萬個現代和已滅絕生物的分子——以發現新的抗生素候選藥物。

研究以「

Deep-learning-enabled antibiotic discovery through molecular de-extinction

」為題，於2024 年6 月11 日發佈於《

Nature Biomedical Engineering》。

論文連結：https://www.nature.com/articles/s41551-024-01201-x

#「採用傳統方法，開發治療感染的新型臨床前候選藥物大約需要六年時間，這個過程非常艱苦且昂貴。

「我們的深度學習方法可以大大縮短時間，降低成本，因為我們在短短幾個小時內就確定了數千種候選藥物，而且其中許多藥物具有臨床前潛力，這在我們的動物模型中進行了測試，標誌著抗生素發現的新時代的開始。該團隊提出了一個基本問題：能否利用機器透過挖掘全球生物資訊來加速抗生素的發現？

他解釋說，這個想法基於這樣的觀念：生物學從最基本的層面上來說是一個資訊來源，理論上可以利用人工智慧進行探索，以尋找新的有用分子。

團隊首先應用簡單的演算法來挖掘單一蛋白質，以找到隱藏在其氨基酸序列中的小抗生素分子。隨著運算能力的進步，De la Fuente 意識到他們可以將挖掘單一蛋白質擴展到挖掘整個蛋白質組。

他說，他們能夠挖掘「整個蛋白質組，即生物體基因組中編碼的所有蛋白質，這使我們在人類蛋白質組中發現了數千種新的抗菌分子，後來又在尼安德特人和丹尼索瓦人等古代類人猿的蛋白質組中發現了數千種新的抗菌分子。他說。

「分子復活」技術

De la Fuente 團隊開發了所謂的「分子復活」技術，即復活已經滅絕的具有潛在治療作用的古代分子，並因此在古代生物的基因組中發現了治療分子。他們推測，他們發現的許多分子可能在整個進化過程中為宿主的免疫發揮作用。

研究以「

Discovery of antimicrobial peptides in the global microbiome with machine learning」為題，於 2024 年 6 月 5 日發佈在《Cell

》。

論文連結：https://doi.org/10.1016/j.cell.2024.05.013

#研究者在《Cell

》的這項研究中提出了一種基於機器學習的方法，來預測全球微生物組中的抗菌肽(AMP)，並利用來自環境和宿主相關棲息地的63,410 個宏基因組和87,920 個原核生物基因組的龐大資料集來建立AMPSphere，這是一個包含863,498 種非冗餘勝肽的綜合目錄，其中很少有與現有資料庫相符的勝肽。

AMPSphere 提供了勝肽演化起源的見解，包括透過複製或較長序列的基因截斷，研究人員觀察到 AMP 的產生因棲息地而異。

為了驗證預測，研究人員合成了 100 種 AMP，並在體外和體內測試了它們對臨床相關的抗藥性病原體和人類腸道共生菌的作用。

A total of 79 peptides were active, 63 of which targeted pathogens. These active AMPs exhibit antimicrobial activity by disrupting bacterial membranes. In total, this approach identified nearly one million prokaryotic AMP sequences, an open source for antibiotic discovery.

antibiotic peptide de-extinction

Illustration: Molecular removal of antibiotics from ancient proteomes using deep learning. (Source: paper)

In the study in "Nature Biomedical Engineering, researchers show that deep learning can be used to mine the proteomes of all available extinct organisms to discover antibiotic peptides.

De la Fuente’s team trained a combination of deep learning models consisting of peptide sequence encoders and neural networks, called Antibiotic Peptide De-Extinction (APEX), to predict antimicrobial activity and used it to mine 10,311,899 peptides .

Marcelo Der Torossian Torres, a postdoctoral researcher in De la Fuente's lab, said that when the team built APEX, they first created a "highly standardized data set to train it, which was missing in the literature... This is surprising because there are so many data sets and researchers will use multiple data sets, assuming that all samples are collected in a very systematic and consistent way, which is not always the case."

APEX did also utilize "probably the largest data set of its kind" as a control for the experiment, he said. This allows researchers to determine how their models perform relative to existing knowledge and validate the uniqueness and validity of antibiotic sequences discovered by APEX.

"Only with high-quality data sets can artificial intelligence succeed in a complex and messy field like biology." De la Fuente said, "We realized this many years ago and have been working hard. Create a data set that can be used to train our algorithm."

APEX uses a combination of recurrent neural networks and attention networks to perform two key tasks, namely identifying encryption," said Fangping Wan, a postdoctoral researcher in De la Fuente's lab. Peptides, fragments within proteins that have antimicrobial properties.

"Recurrent neural networks are very good at processing sequences, such as proteins, because they can process input independent and ordered data." Wan said, "And attention networks can improve the network's localization of proteins that may be related to antibacterial activity. The models predicted 37,176 sequences with broad-spectrum antibacterial activity, 11,035 of which were not found in extant organisms.

Synthesis and Application Validation

They also synthesized 69 peptides and experimentally confirmed their activity against bacterial pathogens. Most peptides kill bacteria by depolarizing their cytoplasmic membrane, in contrast to known antimicrobial peptides, which tend to target the outer membrane.

It is worth noting that some of the lead compounds (including mammothin-2 from mammoths, pixel-2 from straight-tusked elephants, hydrogenated damin-1 from ancient manatees, from giant trees Lazy carnosine-2 and macrocerocin-1 from the extinct giant elk) showed anti-infectious activity in mice with skin abscesses or thigh infections.

This is a crucial step as it brings these drug candidates closer to potential clinical trials and eventual therapeutic use.

In addition, most of the ancient peptides have a novel mechanism of action by depolarizing bacterial cell membranes, a unique targeting approach that suggests a new paradigm for infectious disease control.

Taken together, the computational work performed by De la Fuente’s lab over the past five years has significantly accelerated the ability to discover new antibiotics. What used to take years of hard work using traditional methods can now be done in just a few hours using AI.

Related reports:

https://phys.org/news/2024-06-ai-antibiotic-resistance.html

The above is the detailed content of 'Resurrection' of ancient biological molecules, AI solves antibiotic resistance, two papers published by Fudan University and Penn University collaborative teams were published in Cell and Nature sub-journals. For more information, please follow other related articles on the PHP Chinese website!