Original title: H-Packer: Holographic Rotationally Equivariant Convolutional Neural Network for Protein Side-Chain Packing
Authors: Gian Marco Visani, William Galvin, Michael Neal Pun, Armita Nourmohammad
In this article, the authors discuss the importance of accurately modeling protein 3D structure for the development of functional proteins. They specifically focus on the task of protein side-chain packing, which involves predicting the conformation of side-chains given the backbone structure and amino-acid sequence of the protein. Traditionally, this task has relied on expensive sampling procedures and hand-crafted energy functions and rotamer libraries.
However, recent advancements in deep learning have provided new approaches to tackle this problem. The authors propose a novel algorithm called Holographic Packer (H-Packer), which uses two light-weight rotationally equivariant neural networks. They frame the problem as a joint regression over the side-chains’ dihedral angles, taking into account the underlying symmetries of the task.
The authors evaluate H-Packer using CASP13 and CASP14 targets and find that it is computationally efficient and performs favorably against conventional physics-based algorithms. It also competes well against other deep learning solutions. This research offers a promising method for accurately predicting protein side-chain packing.
Original article: https://arxiv.org/abs/2311.09312