The development of biological drugs (medicines that contain one or more active ingredients produced or extracted from a biological system) is a long and expensive process. Usually, it starts with the study of proteins, and then arrives at the slow and scrupulous process of their transformation into effective and safe drugs. But the advent of artificial intelligence at the service of medicine, which involves the use of advanced analytical-instrumental techniques and new research methods, is changing every aspect. Especially thanks to generative biology.
Generative AI for science
An innovative approach to the development of new protein therapies based on machine learning and artificial intelligence, generative biology allows the development of new biological drugs (the category includes hormones, enzymes, blood products, serums and vaccines, immunoglobulins, allergens , monoclonal antibodies) with the desired structure and properties. A process similar to the one in which generative artificial intelligence systems (such as OpenAI’s ChatGPT) allow you to create new data, texts or images starting from real elements.
Specifically, researchers are starting to use protein-specific data to train machine learning algorithms to design molecules faster and more effectively. Thus, “new” proteins can be evaluated in the laboratory through automated platforms, returning additional information to experts to refine machine learning models in a sort of generative loop.
Design new protein models from scratch
If in the past the study of the structure of a protein represented a complex challenge for computer models, today the latter are even able to design proteins from scratch that do not exist in nature. Consequently, improving drug development. A relevant result achieved by applying AI, in particular through the RosettaFold Diffusion program, whose algorithm is capable of designing proteins by combining structure prediction networks and generative diffusion models.
The new method – published in the journal Nature and perfected by the research team led by David Baker of the University of Washington in Seattle – allows the rapid creation of molecules useful in numerous sectors (such as biomarkers for identifying pollutants or pathologies). “It is a significant step forward compared to what had been learned to do so far”, the comment of Marco Marcia, of the European Laboratory of Molecular Biology, the research institute founded in 1974 which – using the AlphaFold artificial intelligence algorithm developed in collaboration with DeepMind – had made it possible in August 2022 to predict the three-dimensional structure of the majority of proteins in every living organism (over 200 million).