Spherical convolutions on molecular graphs for protein model quality assessment

Abstract Processing information on three-dimensional (3D) objects requires methods stable to rigid-body transformations, in particular rotations, of the input data. In image processing tasks, convolutional neural networks achieve this property using rotation-equivariant operations. However, contrary images, graphs generally have irregular topology. This makes it challenging define a convolution operation these structures. work, we propose spherical graph network that processes 3D models proteins represented as molecular graphs. protein molecule, individual amino acids common topological elements. allows us unambiguously associate each acid with local coordinate system and construct filters operate angular between nodes. Within framework model quality assessment problem, demonstrate proposed method significantly improves compared standard message-passing approach. It is also comparable state-of-the-art methods, critical structure prediction benchmarks. The technique operates only geometric features models. universal applicable any other geometric-learning task where constructing systems. available at https://team.inria.fr/nano-d/software/s-gcn/ .