Brian Ross, MD, PhD, and colleagues in the UNC School of Medicine lab and colleagues at UCSF, Stanford, and Harvard University are showing that protein prediction technology can yield accurate results in the search to efficiently find the best drug candidates for many conditions. I decided that it could be done.
chapel hill, north carolina – Artificial intelligence (AI) has numerous applications in healthcare, from analyzing medical images to optimizing the conduct of clinical trials and even accelerating drug discovery.
AlphaFold2, an artificial intelligence system that predicts protein structures, has enabled scientists to identify and conjure up an almost limitless number of drug candidates for the treatment of neuropsychiatric disorders. However, recent studies have raised questions about AlphaFold2's accuracy in modeling ligand binding sites, the regions on proteins where drugs bind and initiate signaling within cells to cause therapeutic effects, and possible side effects. I am.
In a new paper, Brian Ross, MD, Michael Hooker Distinguished Professor of Pharmacology at the University of North Carolina School of Medicine and director of the NIMH Psychoactive Drug Screening Program, and colleagues at UCSF, Stanford, and Harvard University, describe the AlphaFold2 technology. You can obtain accurate results for the ligand-bound structure even if there are no problems with the ligand binding structure. Their results are science.
“Our results suggest that the AF2 structure may be useful for drug discovery,” said senior author Ross, who also holds an appointment at the UNC Eshelman School of Pharmacy and is a member of the UNC Lineberger Comprehensive Cancer Center. he said. “This type of AI tool could be invaluable, as the possibilities for creating drugs that act on the intended target to treat disease are nearly limitless.”
AlphaFold2 and future modeling
Much like weather forecasts or stock market predictions, AlphaFold2 works by drawing from a vast database of known proteins to create a model of protein structure. You can then simulate how different molecular compounds (such as drug candidates) fit into the protein's binding sites and produce the desired effect. Researchers can use the resulting combinations to better understand protein interactions and create new drug candidates.
To determine AlphaFold2's accuracy, researchers needed to compare the results of a retrospective study with those of a prospective study. Retrospective studies involve researchers giving predictive software compounds that are already known to bind to receptors. Prospective studies, on the other hand, require researchers to use the technology in new conditions and input information about compounds that interact with receptors or not into an AI platform.
The researchers used two proteins in this study: sigma-2 and 5-HT2A. These proteins belong to two distinct protein families and are important in cellular communication and are thought to be involved in neuropsychiatric diseases such as Alzheimer's disease and schizophrenia. The 5-HT2A serotonin receptor is also a major target of psychedelic drugs that hold promise for the treatment of numerous neuropsychiatric disorders.
AlphaFold2 had no prior information about sigma-2 and 5-HT2A or the compounds that might bind to them, so Roth and colleagues chose these proteins. Essentially, the technology is given two untrained proteins, effectively giving researchers a “blank slate.”
First, the researchers entered the protein structures of sigma-2 and 5-HT2A into the AlphaFold system and created a predictive model. Researchers then accessed physical models of his two proteins, which were generated using complex microscopy and X-ray crystallography techniques. At the push of a button, 1.6 billion potential drugs were targeted to the experimental model and his AlphaFold2 model. Interestingly, drug candidates had different results in all models.
success rate
Although the results of the models vary, they hold great promise for drug discovery. The researchers determined that the proportion of compounds that actually changed protein activity in each model was about 50% for the sigma-2 receptor and about 20% for the 5-HT2A receptor. Results above 5% are exceptional.
Out of hundreds of millions of potential combinations, 54% of drug-protein interactions using the sigma-2 AlphaFold2 protein model were successfully activated by the bound drug candidate. Similar results were obtained in the Sigma 2 experimental model with a success rate of 51%.
“This research would not have been possible without the collaboration of several leading experts at UCSF, Stanford University, Harvard University, and UNC-Chapel Hill,” Ross said. “Going forward, we will test whether these results can be applied to other therapeutic targets and target classes.”
Other researchers involved in this study include Dr. Nicholas Kapolka. Dr. Ryan Gamper. Dr. Kensuke Sakamoto; Dr. Yujun Kim. Dr. Jeffrey Dibert. and Dr. Kugure Kim.
Media contact: Kendall DanielsCommunication Specialist, UNC Health | UNC School of Medicine
