Researchers at the University of Toronto’s Donelly Centre have developed a state-of-the-art Artificial Intelligence (AI) model, PepFlow, which accurately predicts the form of peptides. Peptides are smaller molecules made up of amino acids, the essential constituents of proteins. The size and flexibility of peptides allow them to fold into different shapes; the precise shape further dictates the peptide’s interaction with other body molecules and its biological function.
Predicting peptide and protein shapes has long been a challenge in biology. However, AI models like AlphaFold 2 and 3 created by Google’s DeepMind have radically advanced protein structure prediction. Using deep learning, AlphaFold2 predicts the 3D structure of proteins based on their amino acid sequence. Nevertheless, when dealing with enormously flexible molecules like peptides, AlphaFold2 has certain limitations.
PepFlow, as documented in a study in Nature Machine Intelligence, overcomes these limitations by utilizing deep-learning to capture the precise conformation of a peptide within minutes. The AI models used in PepFlow are inspired by Boltzmann generators which help understand the basic physical principles governing a peptide’s chemical structure that influences its spectrum of possible shapes. PepFlow’s ability to predict the complete “energy landscape” of a peptide that represents all possible shapes and transitions between these conformations, sets it apart from models like AlphaFold2.
Peptides play a crucial role in the development of drugs due to their unique binding properties. As a result, PepFlow can have significant implications for developing peptide-based therapeutics. Peptide drugs possess several advantages like specificity in their actions, lower toxicity, and are easier and less expensive to produce compared to larger protein drugs.
In the words of Philip M. Kim, the study’s lead investigator, “Peptides were the focus of the PepFlow model because they are very important biological molecules, and they are naturally very dynamic, so we need to model their different conformations to understand their function.” He added that peptides are immensely important for therapeutics as seen by the GLP1 analogues like Ozempic, used for treating diabetes and obesity. Therefore, PepFlow could expedite the discovery and development of new peptide-based medicines by enabling the design of peptides with therapeutic properties.
The development of PepFlow, which took over two and a half years, followed by one month of training, signals the next frontier, beyond models that predict only one structure of a peptide. The achievement coincides with the release of EvolutionaryScale ESM3, a frontier generative model for biology focusing on proteins.
