FunQG: Molecular Representation Learning Via Quotient Graphs

Research output: Contribution to journalArticleResearchpeer-review

Standard

FunQG: Molecular Representation Learning Via Quotient Graphs. / Hajiabolhassan, Hossein; Taheri, Zahra; Hojatnia, Ali et al.
In: Journal of chemical information and modeling, Vol. 63.2023, No. 11, 15.05.2023, p. 3275-3287.

Research output: Contribution to journalArticleResearchpeer-review

Vancouver

Hajiabolhassan H, Taheri Z, Hojatnia A, Taheri Yeganeh Y. FunQG: Molecular Representation Learning Via Quotient Graphs. Journal of chemical information and modeling. 2023 May 15;63.2023(11):3275-3287. doi: 10.1021/acs.jcim.3c00445

Author

Hajiabolhassan, Hossein ; Taheri, Zahra ; Hojatnia, Ali et al. / FunQG: Molecular Representation Learning Via Quotient Graphs. In: Journal of chemical information and modeling. 2023 ; Vol. 63.2023, No. 11. pp. 3275-3287.

Bibtex - Download

@article{29a29f66de084bc89ebbffd6d1085061,
title = "FunQG: Molecular Representation Learning Via Quotient Graphs",
abstract = "To accurately predict molecular properties, it is important to learn expressive molecular representations. Graph neural networks (GNNs) have made significant advances in this area, but they oftenface limitations like neighbors-explosion, under-reaching, over-smoothing, and over-squashing. Additionally, GNNs tend to have high computational costs due to their large number of parameters.These limitations emerge or increase when dealing with larger graphs or deeper GNN models. Onepotential solution is to simplify the molecular graph into a smaller, richer, and more informative onethat is easier to train GNNs. Our proposed molecular graph coarsening framework called FunQG,uses Functional groups as building blocks to determine a molecule{\textquoteright}s properties, based on a graph-theoretic concept called Quotient Graph. We show through experiments that the resulting informativegraphs are much smaller than the original molecular graphs and are thus more suitable for trainingGNNs. We apply FunQG to popular molecular property prediction benchmarks and compare theperformance of popular baseline GNNs on the resulting datasets to that of state-of-the-art baselineson the original datasets. Our experiments demonstrate that FunQG yields notable results on variousdatasets while dramatically reducing the number of parameters and computational costs. By utilizingfunctional groups, we can achieve an interpretable framework that indicates their significant role indetermining the properties of molecular quotient graphs. Consequently, FunQG is a straightforward,computationally efficient, and generalizable solution for addressing the molecular representationlearning problem.",
author = "Hossein Hajiabolhassan and Zahra Taheri and Ali Hojatnia and {Taheri Yeganeh}, Yavar",
year = "2023",
month = may,
day = "15",
doi = "10.1021/acs.jcim.3c00445",
language = "English",
volume = "63.2023",
pages = "3275--3287",
journal = "Journal of chemical information and modeling",
issn = "1549-9596",
publisher = "American Chemical Society",
number = "11",

}

RIS (suitable for import to EndNote) - Download

TY - JOUR

T1 - FunQG: Molecular Representation Learning Via Quotient Graphs

AU - Hajiabolhassan, Hossein

AU - Taheri, Zahra

AU - Hojatnia, Ali

AU - Taheri Yeganeh, Yavar

PY - 2023/5/15

Y1 - 2023/5/15

N2 - To accurately predict molecular properties, it is important to learn expressive molecular representations. Graph neural networks (GNNs) have made significant advances in this area, but they oftenface limitations like neighbors-explosion, under-reaching, over-smoothing, and over-squashing. Additionally, GNNs tend to have high computational costs due to their large number of parameters.These limitations emerge or increase when dealing with larger graphs or deeper GNN models. Onepotential solution is to simplify the molecular graph into a smaller, richer, and more informative onethat is easier to train GNNs. Our proposed molecular graph coarsening framework called FunQG,uses Functional groups as building blocks to determine a molecule’s properties, based on a graph-theoretic concept called Quotient Graph. We show through experiments that the resulting informativegraphs are much smaller than the original molecular graphs and are thus more suitable for trainingGNNs. We apply FunQG to popular molecular property prediction benchmarks and compare theperformance of popular baseline GNNs on the resulting datasets to that of state-of-the-art baselineson the original datasets. Our experiments demonstrate that FunQG yields notable results on variousdatasets while dramatically reducing the number of parameters and computational costs. By utilizingfunctional groups, we can achieve an interpretable framework that indicates their significant role indetermining the properties of molecular quotient graphs. Consequently, FunQG is a straightforward,computationally efficient, and generalizable solution for addressing the molecular representationlearning problem.

AB - To accurately predict molecular properties, it is important to learn expressive molecular representations. Graph neural networks (GNNs) have made significant advances in this area, but they oftenface limitations like neighbors-explosion, under-reaching, over-smoothing, and over-squashing. Additionally, GNNs tend to have high computational costs due to their large number of parameters.These limitations emerge or increase when dealing with larger graphs or deeper GNN models. Onepotential solution is to simplify the molecular graph into a smaller, richer, and more informative onethat is easier to train GNNs. Our proposed molecular graph coarsening framework called FunQG,uses Functional groups as building blocks to determine a molecule’s properties, based on a graph-theoretic concept called Quotient Graph. We show through experiments that the resulting informativegraphs are much smaller than the original molecular graphs and are thus more suitable for trainingGNNs. We apply FunQG to popular molecular property prediction benchmarks and compare theperformance of popular baseline GNNs on the resulting datasets to that of state-of-the-art baselineson the original datasets. Our experiments demonstrate that FunQG yields notable results on variousdatasets while dramatically reducing the number of parameters and computational costs. By utilizingfunctional groups, we can achieve an interpretable framework that indicates their significant role indetermining the properties of molecular quotient graphs. Consequently, FunQG is a straightforward,computationally efficient, and generalizable solution for addressing the molecular representationlearning problem.

U2 - 10.1021/acs.jcim.3c00445

DO - 10.1021/acs.jcim.3c00445

M3 - Article

VL - 63.2023

SP - 3275

EP - 3287

JO - Journal of chemical information and modeling

JF - Journal of chemical information and modeling

SN - 1549-9596

IS - 11

ER -