New Delhi, Aug. 18 -- Scientists at the Massachusetts Institute of Technology have achieved a breakthrough in the fight against antibiotic-resistant bacteria, using artificial intelligence (AI) to design completely new antibiotics from scratch-molecules that have never before existed in nature or laboratories. The AI-designed drugs successfully treated some infections in laboratory animals, including strains of bacteria that resist all current antibiotics, according to research published on Wednesday in the journal Cell. The work represents the first time that artificial intelligence has been used to create entirely novel antibiotic compounds rather than modify existing ones. Antibiotic resistance has emerged as one of the most pressing global health threats. A major study published in The Lancet in September 2024 estimated more than a million deaths from antibiotic-resistant infections globally each year since 1990, with such fatalities projected to increase by nearly 70 per cent by 2050. The analysis warns that more than 39 million people could die from drug-resistant infections over the next 25 years without decisive action. "We're excited about the new possibilities that this project opens up for antibiotic development," said James Collins, the Termeer professor of Medical Engineering and Science in MIT's Institute for Medical Engineering and Science, who led the research. "Our work shows the power of AI from a drug design standpoint, and enables us to exploit much larger chemical spaces that were previously inaccessible," a statement by the university quoted Collins as saying. The MIT researchers found success with two pathogens: methicillin-resistant Staphylococcus aureus (MRSA) and drug resistant strains of gonorrhoea. The Lancet estimate found that deaths from MRSA more than doubled between 1990 and 2021, from 57,200 to 130,000 deaths annually, and certain that strains of gonorrhoea have become virtually untreatable with existing medications. The AI system analysed more than 45 million molecular fragments, using machine learning algorithms to predict which combinations might kill bacteria while remaining safe for human cells. The system operates on principles similar to modern generative AI, such as ChatGPT or DALL-E, but instead of creating text or images, it generates molecular structures. At its core are Graph Neural Networks, which are specialised neural networks that understand molecules as interconnected graphs-where atoms are nodes and chemical bonds are edges-much like how transformer models in ChatGPT understand relationships between words in sentences. These networks were trained on vast databases of known antibacterial compounds, learning to recognise patterns that distinguish effective antibiotics from inactive molecules, similar to how language models learn patterns in text to predict what comes next. The entire system includes multiple layers of AI-powered filtering and scoring. From this vast computational search, the researchers selected 24 promising compounds for laboratory synthesis and testing. Seven of the 24 compounds-a 29% success rate-showed selective antibacterial activity. By contrast, traditional drug discovery methods typically yield far lower success rates. The most promising compound, designated NG1, proved effective against Neisseria gonorrhoeae, the bacterium that causes gonorrhoea. In laboratory tests, NG1 killed antibiotic-resistant strains of the bacterium and showed efficacy in mouse infection models. Crucially, the compound works through a previously unexploited mechanism, targeting a protein called LptA that helps bacteria build their outer protective membrane. This approach means existing resistance mechanisms are unlikely to protect bacteria from the new drug. "The compound displayed unique modes of action against N gonorrhoeae," the researchers reported in their study. Resistance studies revealed that bacteria struggled to develop defences against NG1, with spontaneous resistance emerging at a frequency of less than one in a billion-exceptionally low for bacterial pathogens. The research team also developed compounds effective against staphylococcus aureus, including MRSA that cause life-threatening infections in hospitals. One compound, EN1, demonstrated activity against both drug-sensitive and drug-resistant strains while showing minimal toxicity to human cells. The AI approach represents a fundamental shift from traditional antibiotic discovery, which typically involves screening natural compounds produced by soil bacteria and fungi. Instead, the MIT team used two complementary computational strategies. The first, called fragment-based design, starts with small molecular building blocks that show antibacterial activity, then uses AI to expand and modify these fragments into full drug molecules. The second approach, termed de novo design, generates completely novel molecular structures without any starting template. "Genetic algorithms and variational auto-encoders enable fragment-based and de novo design," the researchers wrote, describing their computational toolkit. The AI systems learned to recognise patterns associated with antibacterial activity by training on databases of known antimicrobial compounds. The algorithms also incorporated filters to eliminate compounds likely to be toxic to human cells or difficult to synthesise in pharmaceutical laboratories. This computational screening dramatically reduced the number of compounds requiring expensive and time-consuming laboratory testing. While the results represent a significant scientific advance, considerable work remains before AI-designed antibiotics not just to reach patients but prove their safety and efficacy in humans. The most promising compounds must undergo extensive safety testing, followed by clinical trials to establish their effectiveness and safety in humans-a process that typically requires 10 to 15 years and costs hundreds of millions of pounds. Estimates suggest 90% of conventionally discovered drug candidates fail to make the cut. The researchers are now working to optimise their lead compounds, improving their potency and drug-like properties while maintaining their novel mechanisms of action. They are also expanding their computational platform to target additional bacterial pathogens, including those that cause tuberculosis and other major infectious diseases. For the pharmaceutical industry, antibiotic development has been a rocky road due to scientific challenges and poor financial returns compared to drugs for chronic diseases....