A team of researchers at the University of Cambridge has been using artificial intelligence (AI) to expedite the search for new treatments for Parkinson’s disease. The AI-based approach has allowed them to screen millions of potential drug compounds and identify promising candidates ten times faster and 1000 times more cost-effectively than traditional methods.
Parkinson’s is a complex, progressive neurodegenerative disease affecting roughly 6 million people globally, and that number is expected to triple by 2040. As there are currently no reliable treatments that can slow or halt the disease’s progression, innovative solutions are in high demand.
Traditional drug discovery methods, which involve screening large chemical libraries to find potential drug candidates, are often slow, expensive, and unsuccessful. A crucial step in searching for Parkinson’s treatments involves identifying small molecules that can inhibit the aggregation of alpha-synuclein, a protein closely associated with the disease. This process is extremely time-consuming, with identifying a lead candidate for further testing taking months or even years.
To address this issue, the research team, led by Professor Michele Vendruscolo, developed a five-step machine learning approach, which was published in Nature Chemical Biology. The process begins with testing small sets of compounds identified through simulations for their potential to inhibit alpha-synuclein aggregation. The results are used to train a machine learning model to predict the qualities of effective compounds. This trained model is then used to quickly screen a virtual library containing millions of compounds and highlight the ones with the most potential. These selected compounds are then further tested in the lab, and the results are fed back into the model to refine its predictive capabilities. After multiple iterations, the system improved the optimization rate of tested compounds from 4% to over 20%.
The compounds identified by the AI were more potent than any previously found, with some demonstrating promising activity at eight-fold lower doses. The model also identified effective compounds with structures that differed from known ones, adding a level of chemical diversity.
The initial success of this project indicates the potential of AI-first approaches in drug discovery for Parkinson’s and other diseases characterized by protein misfolding and aggregation. As the predictive power of these models should further improve with more data and development, AI methods continue to show promise in reshaping medicine and healthcare. While turning these AI-identified candidates into approved treatments is a long journey, the study has shown how machine learning combined with experimental biology can significantly speed up the early stages of drug discovery.
