Multi-View Graph Neural Networks for Molecular Property Prediction

Abstract

The crux of molecular property prediction is to generate meaningful representations of the molecules. One promising route is to exploit the molecular graph structure through Graph Neural Networks (GNNs). It is well known that both atoms and bonds significantly affect the chemical properties of a molecule, so an expressive model shall be able to exploit both node (atom) and edge (bond) information simultaneously. Guided by this observation, we present Multi-View Graph Neural Network (MV-GNN), a multi-view message passing architecture to enable more accurate predictions of molecular properties. In MV-GNN, we introduce a shared self-attentive readout component and disagreement loss to stabilize the training process. This readout component also renders the whole architecture interpretable. We further boost the expressive power of MV-GNN by proposing a cross-dependent message passing scheme that enhances information communication of the two views, which results in the MV-GNN cross variant. Lastly, we theoretically justify the expressiveness of the two proposed models in terms of distinguishing nonisomorphism graphs. Extensive experiments demonstrate that MV-GNN models achieve remarkably superior performance over the state-of-the-art models on a variety of challenging benchmarks. Meanwhile, visualization results of the node importance are consistent with prior knowledge, which confirms the interpretability power of MV-GNN models.

MoleHD: Ultra-Low-Cost Drug Discovery using Hyperdimensional Computing

“…On the other hand, for GNN architectures as a comparison, extensive pre-training can be necessary, e.g., MolCLR requires around 5 days of pre-training using Nvidia® Quadro RTX™ 6000 as reported in the corresponding literature [37]. GNN models are also harder to implement considering the effort of establish multiple layers with considerable amount of nodes, especially the necessity of back-propagation during training with millions of parameters in total [23].…”

“…Thus, MoleHD only needs to run on commodity CPU and can finish both training and testing on the reported datasets within minutes, while GNN requires around 5 days for just training using Nvidia GPU [37]. (3) smaller memory footprint: MoleHD only needs to store a set of vectors for comparison during inference which is usually less than 10 MB, while SOTA neural networks often need millions of nodes and requires memory in 100MB scale to just store the parameters (e.g., weights and activation values) [23].…”

SPA-GCN: Efficient and Flexible GCN Accelerator with an Application for Graph Similarity Computation

“…Because of the vast application of small graphs, numerous algorithms have been proposed to obtain their information by extracting their features or learning how similar two graphs are [3,11,26,30,33,36,43], especially using GCNs [3,26,33,36,43]. In particular, SimGNN [3] proposed a GCN-based approach to learn a similarity score for such graphs.…”

Adversarial Attack Framework on Graph Embedding Models with Limited Knowledge

“…G RAPH embedding models [1]- [5], which elaborate the expressive power of deep learning on graph-structure data, have achieved remarkable success in various domains, such as drug discovery [6]- [9], social network analysis [10]- [12], computer version [13]- [16], medical imaging [16]- [18], financial surveillance [19], structural role classification [20]- [22] and automated machine learning [23]. Given the increasing popularity and success of these methods, a bunch of recent works have posed the risk of graph embedding models against adversarial attacks, just like what the researchers are anxious for convolutional neural networks [24].…”