Cargando…

Robustness of Local Predictions in Atomistic Machine Learning Models

[Image: see text] Machine learning (ML) models for molecules and materials commonly rely on a decomposition of the global target quantity into local, atom-centered contributions. This approach is convenient from a computational perspective, enabling large-scale ML-driven simulations with a linear-sc...

Descripción completa

Detalles Bibliográficos
Autores principales: Chong, Sanggyu, Grasselli, Federico, Ben Mahmoud, Chiheb, Morrow, Joe D., Deringer, Volker L., Ceriotti, Michele
Formato: Online Artículo Texto
Lenguaje:English
Publicado: American Chemical Society 2023
Acceso en línea:https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10688186/
https://www.ncbi.nlm.nih.gov/pubmed/37948446
http://dx.doi.org/10.1021/acs.jctc.3c00704
_version_ 1785152131346464768
author Chong, Sanggyu
Grasselli, Federico
Ben Mahmoud, Chiheb
Morrow, Joe D.
Deringer, Volker L.
Ceriotti, Michele
author_facet Chong, Sanggyu
Grasselli, Federico
Ben Mahmoud, Chiheb
Morrow, Joe D.
Deringer, Volker L.
Ceriotti, Michele
author_sort Chong, Sanggyu
collection PubMed
description [Image: see text] Machine learning (ML) models for molecules and materials commonly rely on a decomposition of the global target quantity into local, atom-centered contributions. This approach is convenient from a computational perspective, enabling large-scale ML-driven simulations with a linear-scaling cost and also allows for the identification and posthoc interpretation of contributions from individual chemical environments and motifs to complicated macroscopic properties. However, even though practical justifications exist for the local decomposition, only the global quantity is rigorously defined. Thus, when the atom-centered contributions are used, their sensitivity to the training strategy or the model architecture should be carefully considered. To this end, we introduce a quantitative metric, which we call the local prediction rigidity (LPR), that allows one to assess how robust the locally decomposed predictions of ML models are. We investigate the dependence of the LPR on the aspects of model training, particularly the composition of training data set, for a range of different problems from simple toy models to real chemical systems. We present strategies to systematically enhance the LPR, which can be used to improve the robustness, interpretability, and transferability of atomistic ML models.
format Online
Article
Text
id pubmed-10688186
institution National Center for Biotechnology Information
language English
publishDate 2023
publisher American Chemical Society
record_format MEDLINE/PubMed
spelling pubmed-106881862023-12-01 Robustness of Local Predictions in Atomistic Machine Learning Models Chong, Sanggyu Grasselli, Federico Ben Mahmoud, Chiheb Morrow, Joe D. Deringer, Volker L. Ceriotti, Michele J Chem Theory Comput [Image: see text] Machine learning (ML) models for molecules and materials commonly rely on a decomposition of the global target quantity into local, atom-centered contributions. This approach is convenient from a computational perspective, enabling large-scale ML-driven simulations with a linear-scaling cost and also allows for the identification and posthoc interpretation of contributions from individual chemical environments and motifs to complicated macroscopic properties. However, even though practical justifications exist for the local decomposition, only the global quantity is rigorously defined. Thus, when the atom-centered contributions are used, their sensitivity to the training strategy or the model architecture should be carefully considered. To this end, we introduce a quantitative metric, which we call the local prediction rigidity (LPR), that allows one to assess how robust the locally decomposed predictions of ML models are. We investigate the dependence of the LPR on the aspects of model training, particularly the composition of training data set, for a range of different problems from simple toy models to real chemical systems. We present strategies to systematically enhance the LPR, which can be used to improve the robustness, interpretability, and transferability of atomistic ML models. American Chemical Society 2023-11-10 /pmc/articles/PMC10688186/ /pubmed/37948446 http://dx.doi.org/10.1021/acs.jctc.3c00704 Text en © 2023 The Authors. Published by American Chemical Society https://creativecommons.org/licenses/by/4.0/Permits the broadest form of re-use including for commercial purposes, provided that author attribution and integrity are maintained (https://creativecommons.org/licenses/by/4.0/).
spellingShingle Chong, Sanggyu
Grasselli, Federico
Ben Mahmoud, Chiheb
Morrow, Joe D.
Deringer, Volker L.
Ceriotti, Michele
Robustness of Local Predictions in Atomistic Machine Learning Models
title Robustness of Local Predictions in Atomistic Machine Learning Models
title_full Robustness of Local Predictions in Atomistic Machine Learning Models
title_fullStr Robustness of Local Predictions in Atomistic Machine Learning Models
title_full_unstemmed Robustness of Local Predictions in Atomistic Machine Learning Models
title_short Robustness of Local Predictions in Atomistic Machine Learning Models
title_sort robustness of local predictions in atomistic machine learning models
url https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10688186/
https://www.ncbi.nlm.nih.gov/pubmed/37948446
http://dx.doi.org/10.1021/acs.jctc.3c00704
work_keys_str_mv AT chongsanggyu robustnessoflocalpredictionsinatomisticmachinelearningmodels
AT grassellifederico robustnessoflocalpredictionsinatomisticmachinelearningmodels
AT benmahmoudchiheb robustnessoflocalpredictionsinatomisticmachinelearningmodels
AT morrowjoed robustnessoflocalpredictionsinatomisticmachinelearningmodels
AT deringervolkerl robustnessoflocalpredictionsinatomisticmachinelearningmodels
AT ceriottimichele robustnessoflocalpredictionsinatomisticmachinelearningmodels