Scientists have found that a protein prediction technology can effectively identify the most suitable drug candidates for various conditions. The use of artificial intelligence (AI) in healthcare has multiple applications, including analyzing medical images, improving clinical trial processes, and aiding in drug discovery. One specific AI system, AlphaFold2, has enabled scientists to predict protein structures, leading to the discovery of countless potential drug candidates for treating neuropsychiatric disorders.tric disorders. Recent studies have raised concerns about the reliability of AlphaFold2 in predicting ligand binding sites, which are crucial for the effectiveness and potential side effects of drugs within cells. In a new study, Bryan Roth, MD, PhD, and his colleagues from the University of North Carolina School of Medicine, UCSF, Stanford, and Harvard, found that AlphaFold2 can still produce accurate predictions for ligand binding structures, regardless of the specific technology used.The findings were published in the journal Science. “Our findings indicate that AF2 structures may be helpful for drug discovery,” said Roth, a senior author with a joint appointment at the UNC Eshelman School of Pharmacy. “With countless possibilities for developing drugs that target specific diseases, this type of AI tool can be extremely valuable.” AlphaFold2 operates similarly to weather forecasting or stock market prediction, using a large database of known proteins to generate protein structure models. It can then simulate The function of AlphaFold2 is to show how various molecular compounds, such as potential drugs, fit into the binding sites of a protein and create the desired effects. This allows researchers to gain insight into how proteins interact and to develop new drug candidates.
In order to validate the accuracy of AlphaFold2, researchers needed to compare the results of a retrospective study with those of a prospective study. A retrospective study involves researchers inputting compounds that are already known to bind to the receptor into the prediction software. On the other hand, a prospective study requires researchers to use the technology without prior knowledge and then provide the AI platform with information.
The study focused on compounds that may or may not have an effect on the receptor.
For the research, scientists utilized two proteins, sigma-2 and 5-HT2A. These proteins are part of different protein families and play a crucial role in cell communication. They have also been linked to conditions like Alzheimer’s disease and schizophrenia. Additionally, the 5-HT2A serotonin receptor is a key target for psychedelic drugs, which show potential for treating various neuropsychiatric disorders.
Roth and his team chose these proteins because AlphaFold2 did not have any previous knowledge about sigma-2 and 5-HT2A.The technology was given two proteins that it wasn’t trained on, essentially giving the researchers a “blank slate.” Initially, the AlphaFold system was fed the protein structures for sigma-2 and 5-HT2A to create a prediction model. The researchers then accessed physical models of the two proteins produced using complex microscopy and x-ray crystallography techniques. With a press of a button, up to 1.6 billion potential drugs were targeted to the experimental and AlphaFold2 models. Interestingly, each model had a different drug candidate outcome.
High Success Rates
Although the models had varying results, they demonstrate promising potential for drug discovery. Researchers found that about 50% of the compounds were able to alter protein activity for the sigma-2 receptor, and about 20% for the 5-HT2A receptors. Any result exceeding 5% is considered exceptional.
Out of the numerous potential combinations, 54% of the drug-protein interactions using the sigma-2 AlphaFold2 protein models were successfully activated by a bound drug candidate. The experimental model for sigma-2 yielded similar successful results.With a 51% success rate.
“This project would not have been possible without the collaboration of leading experts from UCSF, Stanford, Harvard, and UNC-Chapel Hill,” Roth stated. “Moving forward, we will explore whether these findings can be applied to other therapeutic targets and target classes.”
