Study: AI accelerates drug development and personalized medicine
The creation process for new active ingredients can be made much more efficient with AI, from the idea to approval, explains the Learning Systems platform.
(Image: metamorworks/Shutterstock.com)
Working on new drugs is a complex, lengthy undertaking: It takes up to twelve years to bring them to market – at an average total cost of around 2.8 billion US dollars. The reasons for this are increasingly complex products and study designs, increasing documentation and safety requirements and the costly recruitment of participants for clinical studies. Pharmaceutical companies are therefore reluctant to design new active ingredients such as antibiotics if they are no longer considered profitable. However, systems with artificial intelligence (AI) offer a means of counteracting this for the industry and the healthcare sector as a whole, according to a recent study by the Learning Systems platform.
The process of drug development can be made significantly more efficient "from the initial idea to approval" with the help of AI, writes the team led by Klemens Budde from Charité in the network's white paper, which is hosted by the German Academy of Science and Engineering (Acatech) and funded by the Federal Ministry of Education and Research. This makes it possible to "save years of work and costly investments". "Systematic analyses in data processing", for example to recognize relevant patterns from big data, are crucial for this.
With the help of AI, huge amounts of data can be systematically analyzed and extensive knowledge can be evaluated quickly, the authors explain. This makes it possible to find suitable drug targets and candidates in a short space of time, make better predictions about the side effects of drugs and optimize chemical synthesis, i.e. the production of the active ingredient. The key technology could also help with the selection and monitoring of test subjects for clinical trials and approval. AI-based data analysis also enables the development of personalized therapies for the treatment of cancer, for example. These could be better tailored to the patient's individual clinical picture.
Challenge: filling gaps in the database
The AlphaFold software developed by Google DeepMind enables the AI-based prediction of crucial protein structures "within a few hours with high accuracy", the members of the platform's Health, Medical Technology and Care working group give an example. In order to achieve comparable accuracy and resolution, such molecule strands would previously have had to be researched experimentally, sometimes over months. The US biotechnology company Insilico Medicine was also able to develop a drug candidate against fibrosis through to the preclinical phase with AI support for less than USD 850,000. Traditionally, this would have cost around 664 million US dollars.
According to the analysis, the South Korean pharmaceutical tech company Standigm has also developed an AI-based platform for identifying drugs with new mechanisms of action, which allows these structures to be identified within an average of seven months compared to the typical 30 months. The authors also discuss the opportunities offered by generative AI. Med-PaLM, for example, is a language model developed by Google specifically for medical questions. It supports "the intuitive, text-based query of relevant genes for certain diseases based on information organized in knowledge graphs". As a counterpart, Exscientia has published a chatbot for producing knowledge graphs. Generative AI can also be used to create new molecules or proteins. However, hallucinations of such language models are challenging.
The experts cite the lack of legal requirements and the quality and availability of data as general hurdles. The willingness of research-based companies to share information is important. Gaps exist in particular in the database on human biology, for example on disease mechanisms and the effect of drugs. These could be closed with high-quality measurements from the population, which would ideally be made available via the electronic patient record (ePA) or health insurance companies. Politicians must set the right course, for example with the European Health Data Space (EHDS) and standards such as the Health Data Utilization Act. However, AI-supported research should not be torpedoed by barriers such as those demanded by civil rights activists.
(jren)
