Published on Fri Aug 02 2019

Universal Transforming Geometric Network

Jin Li

The recurrent geometric network (RGN) is the first end-to-end differentiable neural architecture for protein structure prediction. RGN's use of recurrent neural networks (RNNs) as internal representations results in long training time and unstable gradients. We propose the Universal

0
0
0
Abstract

The recurrent geometric network (RGN), the first end-to-end differentiable neural architecture for protein structure prediction, is a competitive alternative to existing models. However, the RGN's use of recurrent neural networks (RNNs) as internal representations results in long training time and unstable gradients. And because of its sequential nature, it is less effective at learning global dependencies among amino acids than existing transformer architectures. We propose the Universal Transforming Geometric Network (UTGN), an end-to-end differentiable model that uses the encoder portion of the Universal Transformer architecture as an alternative for internal representations. Our experiments show that compared to RGN, UTGN achieve a \si{\angstrom} improvement on the free modeling portion and a \si{\angstrom} improvement on the template based modeling of the CASP12 competition.

Sun May 16 2021
Machine Learning
Protein sequence-to-structure learning: Is this the end(-to-end revolution)?
The potential of deep learning has been recognized in the protein structure predicting community for some time, and became indisputable after CASP13. Here, we provide an overview and our opinion of the novel deep learning approaches developed in the last two years and widely used in CASP14.
7
78
328
Tue Jun 22 2021
Computer Vision
G-VAE, a Geometric Convolutional VAE for ProteinStructure Generation
Analyzing the structure of proteins is a key part of understanding their functions. In this work, we introduce a joint geometric-neural networks approach for comparing, deforming and generating 3D structures.
1
0
0
Fri Nov 23 2018
Machine Learning
TorchProteinLibrary: A computationally efficient, differentiable representation of protein structure
Predicting the structure of a protein from its sequence is a cornerstone task of molecular biology. In this paper we present a library of differentiable mappings from two standardhedral-angle representations of protein structure to atomic Cartesian coordinates.
0
0
0
Fri Oct 16 2020
Machine Learning
SidechainNet: An All-Atom Protein Structure Dataset for Machine Learning
SidechainNet is a new dataset that directly extends the ProteinNet dataset. It includes angle and atomiccoordinate information capable of describing all heavy atoms of each protein.
0
0
0
Mon Oct 29 2018
Machine Learning
DeepSymmetry : Using 3D convolutional networks for identification of tandem repeats and internal symmetries in protein structures
DeepSymmetry is a versatile method based on three-dimensional convolutional networks. It can detect proteins with structural repetitions or internal symmetries. The method is available at https://team.inria.fr/nano-d/software/deepsymmetry.
0
0
0
Tue May 10 2016
Machine Learning
MUST-CNN: A Multilayer Shift-and-Stitch Deep Convolutional Architecture for Sequence-based Protein Structure Prediction
Predicting protein properties is an important task in bioinformatics. We propose a deep convolutional neural network to predict protein properties. We beat the state-of-the-art performance on two large protein property prediction datasets.
0
0
0