(197e) Deep Learning Architecture for Peptide Property Prediction

Peptides have a wide range of applications and predicting their properties is important for optimizing their use in various fields such as separation materials for chromatography. Existing approaches to peptide property prediction have limitations and do not scale well to larger molecules or take advantage of important structural features of peptides.In this project, we develop a new deep learning architecture that captures important structural features of peptides, such as hydrogen bonding motifs, to predict their properties. We curate a high-quality dataset of peptide structures and their properties for training and validation of the model, we also evaluate the performance of the model and compare it to existing methods to identify key structural features that contribute to peptide properties. Our method has the potential to significantly advance the design of peptides with optimized properties for a range of applications, future works includes expanding the scope of the model to predict other complex peptide properties and improving the accuracy of the predictions.