ComENet: Towards Complete and Efficient Message Passing for 3D Molecular Graphs

Wang, Limei; Liu, Yi; Lin, Yu‐Hsien; Liu, Haoran; Ji, Shuiwang

doi:10.48550/arxiv.2206.08515

Cited by 6 publications

(6 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Complete Message Passing Mechanism ComENet [46] proposes rotation angles and spherical coordinates to fulfil the global completeness of 3D information on molecular graphs. By incorporating these designed geometric representations into the message passing scheme [18], the complete representation for a whole 3D graph is eventually yielded [47].…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Learning Complete Protein Representation by Deep Coupling of Sequence and Structure

Tan

Xia

et al. 2023

Preprint

View full text Add to dashboard Cite

Learning effective representations is crucial for understanding proteins and their biological functions. Recent advancements in language models and graph neural networks have enabled protein models to leverage primary or tertiary structure information to learn representations. However, the lack of practical methods to deeply co-model the relationships between protein sequences and structures has led to suboptimal embeddings. In this work, we propose CoupleNet, a network that couples protein sequence and structure to obtain informative protein representations. CoupleNet incorporates multiple levels of features in proteins, including the residue identities and positions for sequences, as well as geometric representations for tertiary structures. We construct two types of graphs to model the extracted sequential features and structural geometries, achieving completeness on these graphs, respectively, and perform convolution on nodes and edges simultaneously to obtain superior embeddings. Experimental results on a range of tasks, such as protein fold classification and function prediction, demonstrate that our proposed model outperforms the state-of-the-art methods by large margins.

show abstract

Section: Related Workmentioning

confidence: 99%

“…The operation T g would not change the 3D conformation of a 3D graph [46]. Positions can generate geometric representations, which can also be recovered from them.…”

Section: Given the Functionmentioning

confidence: 99%

Learning Complete Protein Representation by Deep Coupling of Sequence and Structure

Tan

Xia

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…We complement this heuristic scoring method with a quantitative evaluation of the marginal contribution of atomic degrees of freedom to electronic prediction accuracy using equivariant graph neural networks (GNNs). 36 The GNN is chosen as a data-driven metric for identifying the crucial atoms because it can incorporate both the molecular topology and the threedimensional conformation during the learning process (see the SI for details). A training set of 12,000 molecular configurations were sampled via molecular dynamics (MD) simulations, with their electronic structure (highest occupied molecular orbital (HOMO) energy) characterized using density functional theory (DFT).…”

Section: Physically Motivated Scoring Metricmentioning

confidence: 99%

Identifying Coarse-Grained Representations for Electronic Predictions

Wang

Maier

Jackson

2023

J. Chem. Theory Comput.

View full text Add to dashboard Cite

Coarse-grained (CG) simulations are an important computational tool in chemistry and materials science. Recently, systematic “bottom-up” CG models have been introduced to capture electronic structure variations of molecules and polymers at the CG resolution. However, the performance of these models is limited by the ability to select reduced representations that preserve electronic structure information, which remains a challenge. We propose two methods for (i) identifying important electronically coupled atomic degrees of freedom and (ii) scoring the efficacy of CG representations used in conjunction with CG electronic predictions. The first method is a physically motivated approach that incorporates nuclear vibrations and electronic structure derived from simple quantum chemical calculations. We complement this physically motivated approach with a machine learning technique based on the marginal contribution of nuclear degrees of freedom to electronic prediction accuracy using an equivariant graph neural network. By integrating these two approaches, we can both identify critical electronically coupled atomic coordinates and score the efficacy of arbitrary CG representations for making electronic predictions. We leverage this capability to make a connection between optimized CG representations and the future potential for “bottom-up” development of simplified model Hamiltonians incorporating nonlinear vibrational modes.

show abstract

“…Numerous studies recently applied graph neural networks (GNNs) to molecule-related tasks, given the intrinsic graph nature of molecules [3][4][5][6][7][8][9]. GNNs can learn molecular representations Graphical Abstract from molecular input data.…”

Section: Introductionmentioning

confidence: 99%

“…Even though previous studies have some promising results in GNN performance in molecular tasks, many of them either do not compare them with traditional approaches (e.g. [5,12,13] only compares their method with other GNN methods), or the task is not on QSAR modeling (such as quantum mechanics dataset QM9 used in [9]). Recently, some authors expressed doubts about deep learning performance over traditional methods in molecular tasks [14,15].…”

Section: Introductionmentioning

confidence: 99%

Advancements in Ligand-Based Virtual Screening through the Synergistic Integration of Graph Neural Networks and Expert-Crafted Descriptors

Liu

Moretti

Wang

et al. 2023

Preprint

View full text Add to dashboard Cite

In recent years several applications of graph neural networks (GNNs) to molecular tasks have emerged. Whether GNNs outperform the traditional descriptor-based methods in the quantitative structure activity relationship (QSAR) modeling in early computer-aided drug discovery (CADD) remains an open question. This paper introduces a simple yet effective strategy to boost the predictive power of QSAR deep learning models. The strategy proposes to train GNNs together with traditional descriptors, combining the strengths of both methods. The enhanced model consistently outperforms vanilla descriptors or GNN methods on nine well-curated high throughput screening datasets over diverse therapeutic targets.

show abstract

ComENet: Towards Complete and Efficient Message Passing for 3D Molecular Graphs

Cited by 6 publications

References 24 publications

Learning Complete Protein Representation by Deep Coupling of Sequence and Structure

Learning Complete Protein Representation by Deep Coupling of Sequence and Structure

Identifying Coarse-Grained Representations for Electronic Predictions

Advancements in Ligand-Based Virtual Screening through the Synergistic Integration of Graph Neural Networks and Expert-Crafted Descriptors

Contact Info

Product

Resources

About