Intrinsic-Extrinsic Convolution and Pooling for Learning on 3D Protein Structures

Abstract

Proteins perform a large variety of functions in living organisms, thus playing a key role in biology. As of now, available learning algorithms to process protein data do not consider several particularities of such data and/or do not scale well for large protein conformations. To fill this gap, we propose two new learning operations enabling deep 3D analysis of large-scale protein data. First, we introduce a novel convolution operator which considers both, the intrinsic (invariant under protein folding) as well as extrinsic (invariant under bonding) structure, by using n-D convolutions defined on both the Euclidean distance, as well as multiple geodesic distances between atoms in a multi-graph. Second, we enable a multi-scale protein analysis by introducing hierarchical pooling operators, exploiting the fact that proteins are a recombination of a finite set of amino acids, which can be pooled using shared pooling matrices. Lastly, we evaluate the accuracy of our algorithms on several large-scale data sets for common protein analysis tasks, where we outperform state-of-the-art methods.Preprint. Under review.

Pre-training of Equivariant Graph Matching Networks with Conformation Flexibility for Drug Binding

“…Molformer [95] is a variant of Transformer [90] and operates on 3D heterogeneous molecular graphs with motifs. IEConv [35] designs a convolution operator that considers the primary, secondary, and tertiary structure of proteins and a set of hierarchical pooling operators for multi-scale modeling. 3DCNN and 3DGCN [87] are also competitive 3D methods.…”

“…Apart from that, Zhang et al [98] combines a multi-view contrastive learning and a self-prediction learning to encode geometric features of proteins. Then these semantic representations learned from SSL are utilized for downstream tasks including structure classification [35], model quality assessment [4], and function prediction [30]. Nevertheless, no preceding research excavate the potential of pre-training on this sort of spatial-temporal data, partly because of the high expenditure to run MD simulations.…”

Directed Weight Neural Networks for Protein Structure Representation Learning

“…Amidi et al (2018) employs a similar idea to classify enzymes classes by using 3D CNN. 3D CNNs also shed light on other tasks such as interface prediction (Townshend et al, 2019) and protein fold recognition (Hermosilla et al, 2020). Gainza et al (2020); Sverrisson et al (2021) extend 3D CNNs to spherical convolutions for operating on radius regions, which can also be naturally applied to the Fourier space (Zhemchuzhnikov et al, 2021) and the 3D Voronoi Tessellation space (Igashov et al, 2021).…”

“…Therefore, processing such 3D structures is the key for protein function analysis. While we have witnessed remarkable progress in protein structure predictions (Rohl et al, 2004;Källberg et al, 2012;Baek et al, 2021;Jumper et al, 2021), another thread of tasks with protein 3D structures as input starts to draw a great interest, such as function prediction (Hermosilla et al, 2020;Gligorijević et al, 2021), decoy ranking (Lundström et al, 2001;Kwon et al, 2021;Wang et al, 2021), protein docking (Duhovny et al, 2002;Shulman-Peleg et al, 2004;Gainza et al, 2020;Sverrisson et al, 2021), and driver mutation identification (Lefèvre et al, 1997;Antikainen & Martin, 2005;Li et al, 2020;Jankauskaitė et al, 2019).…”

EGR: Equivariant Graph Refinement and Assessment of 3D Protein Complex Structures

“…For example, representing proteins by their amino acid sequences has been shown to provide powerful structural information by comparing sequences to each other [14] and extracting rich unsupervised learning representations using self-attention Transformers [20]. From a geometric viewpoint, several works have aimed to encode protein structural priors directly within neural network architectures to model proteins hierarchically [21,22], as computationally-efficient point clouds [23,24], or as k-nearest neighbors (k-NN) geometric graphs [25,26] for tasks such as protein function prediction [27], protein model quality assessment [28], and protein interaction region prediction [29].…”