Journal article 370 views
Domain-informed graph neural networks: A quantum chemistry case study
Neural Networks, Volume: 165, Pages: 938 - 952
Swansea University Authors:
Jay Morgan, Adeline Paiement, Christian Klinke
Full text not available from this repository: check for access using links below.
DOI (Published version): 10.1016/j.neunet.2023.06.030
Abstract
We explore different strategies to integrate prior domain knowledge into the design of graph neural networks (GNN). Our study is supported by a use-case of estimating the potential energy of chemical systems (molecules and crystals) represented as graphs. We integrate two elements of domain knowledg...
Published in: | Neural Networks |
---|---|
ISSN: | 0893-6080 |
Published: |
Elsevier BV
2023
|
Online Access: |
Check full text
|
URI: | https://cronfa.swan.ac.uk/Record/cronfa63771 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
first_indexed |
2023-07-04T09:35:26Z |
---|---|
last_indexed |
2023-07-04T09:35:26Z |
id |
cronfa63771 |
recordtype |
SURis |
fullrecord |
<?xml version="1.0" encoding="utf-8"?><rfc1807 xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xmlns:xsd="http://www.w3.org/2001/XMLSchema"><bib-version>v2</bib-version><id>63771</id><entry>2023-07-04</entry><title>Domain-informed graph neural networks: A quantum chemistry case study</title><swanseaauthors><author><sid>df9a27bcf77b4769c2ebbb702b587491</sid><firstname>Jay</firstname><surname>Morgan</surname><name>Jay Morgan</name><active>true</active><ethesisStudent>false</ethesisStudent></author><author><sid>f50adf4186d930e3a2a0f9a6d643cf53</sid><firstname>Adeline</firstname><surname>Paiement</surname><name>Adeline Paiement</name><active>true</active><ethesisStudent>false</ethesisStudent></author><author><sid>c10c44238eabfb203111f88a965f5372</sid><ORCID>0000-0001-8558-7389</ORCID><firstname>Christian</firstname><surname>Klinke</surname><name>Christian Klinke</name><active>true</active><ethesisStudent>false</ethesisStudent></author></swanseaauthors><date>2023-07-04</date><deptcode>CRIM</deptcode><abstract>We explore different strategies to integrate prior domain knowledge into the design of graph neural networks (GNN). Our study is supported by a use-case of estimating the potential energy of chemical systems (molecules and crystals) represented as graphs. We integrate two elements of domain knowledge into the design of the GNN to constrain and regularise its learning, towards higher accuracy and generalisation. First, knowledge on the existence of different types of relations/graph edges (e.g. chemical bonds in our case study) between nodes of the graph is used to modulate their interactions. We formulate and compare two strategies, namely specialised message production and specialised update of internal states. Second, knowledge of the relevance of some physical quantities is used to constrain the learnt features towards a higher physical relevance using a simple multi-task learning (MTL) paradigm. We explore the potential of MTL to better capture the underlying mechanisms behind the studied phenomenon. We demonstrate the general applicability of our two knowledge integrations by applying them to three architectures that rely on different mechanisms to propagate information between nodes and to update node states. Our implementations are made publicly available. To support these experiments, we release three new datasets of out-of-equilibrium molecules and crystals of various complexities.</abstract><type>Journal Article</type><journal>Neural Networks</journal><volume>165</volume><journalNumber/><paginationStart>938</paginationStart><paginationEnd>952</paginationEnd><publisher>Elsevier BV</publisher><placeOfPublication/><isbnPrint/><isbnElectronic/><issnPrint>0893-6080</issnPrint><issnElectronic/><keywords>Graph neural network, Domain knowledge integration, Quantum chemistry application</keywords><publishedDay>31</publishedDay><publishedMonth>8</publishedMonth><publishedYear>2023</publishedYear><publishedDate>2023-08-31</publishedDate><doi>10.1016/j.neunet.2023.06.030</doi><url>http://dx.doi.org/10.1016/j.neunet.2023.06.030</url><notes/><college>COLLEGE NANME</college><department>Criminology</department><CollegeCode>COLLEGE CODE</CollegeCode><DepartmentCode>CRIM</DepartmentCode><institution>Swansea University</institution><apcterm/><funders/><projectreference/><lastEdited>2023-11-15T16:46:00.7905532</lastEdited><Created>2023-07-04T10:30:46.7161905</Created><path><level id="1">Faculty of Science and Engineering</level><level id="2">School of Engineering and Applied Sciences - Chemistry</level></path><authors><author><firstname>Jay</firstname><surname>Morgan</surname><order>1</order></author><author><firstname>Adeline</firstname><surname>Paiement</surname><order>2</order></author><author><firstname>Christian</firstname><surname>Klinke</surname><orcid>0000-0001-8558-7389</orcid><order>3</order></author></authors><documents><document><filename>Under embargo</filename><originalFilename>Under embargo</originalFilename><uploaded>2023-07-04T10:34:58.9221881</uploaded><type>Output</type><contentLength>1480288</contentLength><contentType>application/pdf</contentType><version>Proof</version><cronfaStatus>true</cronfaStatus><embargoDate>2024-07-01T00:00:00.0000000</embargoDate><copyrightCorrect>false</copyrightCorrect><language>eng</language></document></documents><OutputDurs/></rfc1807> |
spelling |
v2 63771 2023-07-04 Domain-informed graph neural networks: A quantum chemistry case study df9a27bcf77b4769c2ebbb702b587491 Jay Morgan Jay Morgan true false f50adf4186d930e3a2a0f9a6d643cf53 Adeline Paiement Adeline Paiement true false c10c44238eabfb203111f88a965f5372 0000-0001-8558-7389 Christian Klinke Christian Klinke true false 2023-07-04 CRIM We explore different strategies to integrate prior domain knowledge into the design of graph neural networks (GNN). Our study is supported by a use-case of estimating the potential energy of chemical systems (molecules and crystals) represented as graphs. We integrate two elements of domain knowledge into the design of the GNN to constrain and regularise its learning, towards higher accuracy and generalisation. First, knowledge on the existence of different types of relations/graph edges (e.g. chemical bonds in our case study) between nodes of the graph is used to modulate their interactions. We formulate and compare two strategies, namely specialised message production and specialised update of internal states. Second, knowledge of the relevance of some physical quantities is used to constrain the learnt features towards a higher physical relevance using a simple multi-task learning (MTL) paradigm. We explore the potential of MTL to better capture the underlying mechanisms behind the studied phenomenon. We demonstrate the general applicability of our two knowledge integrations by applying them to three architectures that rely on different mechanisms to propagate information between nodes and to update node states. Our implementations are made publicly available. To support these experiments, we release three new datasets of out-of-equilibrium molecules and crystals of various complexities. Journal Article Neural Networks 165 938 952 Elsevier BV 0893-6080 Graph neural network, Domain knowledge integration, Quantum chemistry application 31 8 2023 2023-08-31 10.1016/j.neunet.2023.06.030 http://dx.doi.org/10.1016/j.neunet.2023.06.030 COLLEGE NANME Criminology COLLEGE CODE CRIM Swansea University 2023-11-15T16:46:00.7905532 2023-07-04T10:30:46.7161905 Faculty of Science and Engineering School of Engineering and Applied Sciences - Chemistry Jay Morgan 1 Adeline Paiement 2 Christian Klinke 0000-0001-8558-7389 3 Under embargo Under embargo 2023-07-04T10:34:58.9221881 Output 1480288 application/pdf Proof true 2024-07-01T00:00:00.0000000 false eng |
title |
Domain-informed graph neural networks: A quantum chemistry case study |
spellingShingle |
Domain-informed graph neural networks: A quantum chemistry case study Jay Morgan Adeline Paiement Christian Klinke |
title_short |
Domain-informed graph neural networks: A quantum chemistry case study |
title_full |
Domain-informed graph neural networks: A quantum chemistry case study |
title_fullStr |
Domain-informed graph neural networks: A quantum chemistry case study |
title_full_unstemmed |
Domain-informed graph neural networks: A quantum chemistry case study |
title_sort |
Domain-informed graph neural networks: A quantum chemistry case study |
author_id_str_mv |
df9a27bcf77b4769c2ebbb702b587491 f50adf4186d930e3a2a0f9a6d643cf53 c10c44238eabfb203111f88a965f5372 |
author_id_fullname_str_mv |
df9a27bcf77b4769c2ebbb702b587491_***_Jay Morgan f50adf4186d930e3a2a0f9a6d643cf53_***_Adeline Paiement c10c44238eabfb203111f88a965f5372_***_Christian Klinke |
author |
Jay Morgan Adeline Paiement Christian Klinke |
author2 |
Jay Morgan Adeline Paiement Christian Klinke |
format |
Journal article |
container_title |
Neural Networks |
container_volume |
165 |
container_start_page |
938 |
publishDate |
2023 |
institution |
Swansea University |
issn |
0893-6080 |
doi_str_mv |
10.1016/j.neunet.2023.06.030 |
publisher |
Elsevier BV |
college_str |
Faculty of Science and Engineering |
hierarchytype |
|
hierarchy_top_id |
facultyofscienceandengineering |
hierarchy_top_title |
Faculty of Science and Engineering |
hierarchy_parent_id |
facultyofscienceandengineering |
hierarchy_parent_title |
Faculty of Science and Engineering |
department_str |
School of Engineering and Applied Sciences - Chemistry{{{_:::_}}}Faculty of Science and Engineering{{{_:::_}}}School of Engineering and Applied Sciences - Chemistry |
url |
http://dx.doi.org/10.1016/j.neunet.2023.06.030 |
document_store_str |
0 |
active_str |
0 |
description |
We explore different strategies to integrate prior domain knowledge into the design of graph neural networks (GNN). Our study is supported by a use-case of estimating the potential energy of chemical systems (molecules and crystals) represented as graphs. We integrate two elements of domain knowledge into the design of the GNN to constrain and regularise its learning, towards higher accuracy and generalisation. First, knowledge on the existence of different types of relations/graph edges (e.g. chemical bonds in our case study) between nodes of the graph is used to modulate their interactions. We formulate and compare two strategies, namely specialised message production and specialised update of internal states. Second, knowledge of the relevance of some physical quantities is used to constrain the learnt features towards a higher physical relevance using a simple multi-task learning (MTL) paradigm. We explore the potential of MTL to better capture the underlying mechanisms behind the studied phenomenon. We demonstrate the general applicability of our two knowledge integrations by applying them to three architectures that rely on different mechanisms to propagate information between nodes and to update node states. Our implementations are made publicly available. To support these experiments, we release three new datasets of out-of-equilibrium molecules and crystals of various complexities. |
published_date |
2023-08-31T16:46:05Z |
_version_ |
1782649208229068800 |
score |
11.016593 |