Preeti
Researchers at the University of Toronto have developed a deep-learning model called PepFlow, which can predict all possible shapes of peptides, smaller chains of amino acids that perform biological functions similar to proteins. The model combines machine learning and physics to model the folding patterns of peptides based on their energy landscape. This innovation has the potential to inform drug development by designing peptides that can act as binders.
Published in Nature Machine Intelligence, the study highlights the importance of understanding peptide folding as it directly impacts their role in the human body and their interactions with other molecules. Peptides are flexible molecules with a wide range of folding patterns, making them crucial in various biological processes and therapeutic development.
PepFlow outperforms the leading Google Deepmind AI system, AlphaFold, by generating a range of conformations for peptides, including unusual formations like macrocycles. The model is inspired by Boltzmann generators and can be easily expanded to incorporate new information and considerations. While PepFlow is the first version of the model and has limitations, it shows promise in advancing treatments that rely on peptide binding to regulate biological processes.
The development of PepFlow took two-and-a-half years and one month to train, showcasing its potential to move beyond predicting a single structure of a peptide. By offering insight into the real energy landscape of peptides, PepFlow opens up new possibilities for drug development and therapeutic innovations.
Source link
Source link: https://phys.org/news/2024-06-deep-outperforms-google-ai-peptide.html
GIPHY App Key not set. Please check settings