AI accelerates drug discovery, but its prospects still need to be tested in practice

　　◎Our reporter Zhang Jiaxin

　　This is perhaps the most dramatic transformation in healthcare: digital biology and generative artificial intelligence (AI) are helping to reshape the drug discovery process.

　　The use of AI to develop new drugs is still in its infancy, but AI-designed drugs have entered the early stages of clinical trials in the past few years, and some AI pharmaceutical pioneer companies have achieved certain results in this field. However, an article published on the British "Nature" website stated that the potential of AI to accelerate drug discovery still needs to be tested in practice.

　　Developing phage forms of antibiotics

　　The human body is occupied by a large number of microorganisms, including viruses. These virus groups are collectively called the human virome. Stefan N. Lukanov, CEO of Salve Therapeutics, an American AI pharmaceutical company, pointed out that naturally occurring viruses in human tissues are an ideal way to carry gene therapy payloads to treat diseases.

　　Salve is combining machine learning with computer-aided design to develop antibiotics in the form of bacteriophages. This approach enables a virtual assessment of the attributes, outcomes and risks of a pharmaceutical invention through extensive iterative analysis of various models.

　　Lukanov said they are working on genetically engineering the phages for greater potency and host range. He expects phage antibiotics to improve the lives of transplant, burn and immune-compromised patients.

　　Lukanov emphasized that because the phages only target bacteria, the antibiotic does not pose significant risks to patients, other than a slight immune response in the body due to the presence of foreign particles.

　　Developing oral small molecule drugs

　　Biolexis Therapeutics, an American AI drug research and development company, specializes in developing oral small molecule drugs for cancer and various metabolic, inflammatory and neurodegenerative diseases.

　　The company discovers and develops new clinical drug candidates through its proprietary MolecuLern process. The process can target any kind of protein, identify new chemical entities with drug-like characteristics and validate them with laboratory data, shortening the time to discover and develop new drugs from years to months. A drug they developed, SLX-0528, is currently in Phase IB trials for pancreatic cancer. The drug is designed to control the cellular differentiation, function and interleukin release of helper T cells 17.

　　Launch of generative AI drug discovery platform

　　Anthony Costa is global head of developer relations for life sciences at NVIDIA. He noted that much generative AI is built on underlying models of large language models. These models are improving their ability to predict drug properties and interactions.

　　To help realize this potential, NVIDIA has developed BioNeMo, a cloud service for generative AI in biology that provides a variety of AI models for small molecules and proteins. Costa asserted that with BioNeMo, researchers can use AI models with proprietary data to quickly predict the 3D structure and function of proteins and biomolecules, which will accelerate the generation of new drug candidates.

　　Chicago-based startup Evozyne recently used BioNeMo to design new proteins to treat phenylketonuria. Phenylketonuria is a rare disorder characterized by elevated levels of the amino acid phenylalanine. Laboratory tests have conclusively proven that some AI-developed protein variants are more effective than their natural forms.

　　AI drug discovery needs clinical validation

　　Drug development involves several specific steps. It typically begins by identifying a biological target that causes a disease (which may include DNA, RNA, protein receptors, or enzymes) and then screening for molecules that may interact with it. This is called the "discovery" phase.

　　New medicines must be rigorous, safe, effective and trustworthy, and companies must find the right path to get there. Even if AI does reduce the time and cost required for compounds to enter preclinical testing, most drug candidates will still fail in later stages. But anything that can speed up the process is a win. Industry and academia must leverage each other’s strengths to determine how to best leverage AI.

　　Lukanov said AI and machine learning represent an exciting new way to improve efficacy and safety and bring more drugs to market. He noted that the use of AI and machine learning in drug discovery is still in its early stages and should be subject to laboratory validation to ensure that only the best drug candidates enter clinical trials.

　　Additionally, various safety features are being incorporated into AI-based drug development. Biolexis, for example, uses multiple methods to prioritize molecules with high safety profile. David J. Beers, the company's chief executive, said the safety and potential unintended consequences of molecules developed with machine learning are important issues that need to be addressed. (Science and Technology Daily)