Visual-PepBDB is a deep learning model that predicts peptide-protein binding sites using Convolutional Neural Networks (CNNs) from Wardaah et al. (2019).

• The model is trained on the PepBDB dataset, which contains 3D structures of peptide-protein complexes.
• Input a 3D structure of a peptide or protein, performs feature enrichment and extraction, and predicts the binding sites.
• Current implementation is available on Deepnote, as hosted here.

With a training accuracy of 78.5% and a sensitivity of 79.3%, Visual-PepBDB was able to outperform the original model's sensitivity by 12 pts.