Welcome to SpatConv!

        Protein-protein interactions (PPIs) play a pivotal role in a multitude of biological processes, with aberrant activities being closely associated with a range of diseases, including cancer and neurodegenerative disorders. This invention proposes an innovative geometric deep learning framework, SpatConv, which directly leverages the raw three-dimensional (3D) coordinates of proteins and features extracted from the pre-trained ProtT5 model as inputs, thereby eliminating the need for any precomputed manual features. SpatConv constructs local coordinate systems that adhere to SE(3) equivariance, enabling precise mapping of residue distribution and integration of multi-scale protein features. The use of spatial graph convolution techniques facilitates efficient message passing and feature updating.

        We provide a convenient and efficient SpatConv webserver for researchers to predict protein binding sites. You can also access the source code of SpatConv on our Github page https://github.com/gmnnnhh/SpatConv.

If you think SpatConv is useful, please kindly cite the following paper:

Webserver update: