Deep Neural Networks for Multicomponent Molecular Systems

Abstract

Deep neural networks (DNNs) represent promising approaches to molecular machine learning (ML). However, their applicability remains limited to single-component materials and a general DNN model capable of handling various multicomponent molecular systems with composition data is still elusive, while current ML approaches for multicomponent molecular systems are still molecular descriptor-based. Here, a general DNN architecture extending existing molecular DNN models to multicomponent systems called MEIA is proposed. Case studies showed that the MEIA architecture could extend two exiting molecular DNN models to multicomponent systems with the same procedure, and that the obtained models that could learn both the molecular structure and composition information with equal or better accuracies compared to a well-used molecular descriptor-based model in the best model for each case study. Furthermore, the case studies also showed that, for ML tasks where the molecular structure information plays a minor role, the performance improvements by DNN models were small; while for ML tasks where the molecular structure information plays a major role, the performance improvements by DNN models were large, and DNN models showed notable predictive accuracies for an extremely sparse dataset, which cannot be modeled without the molecular structure information. The enhanced predictive ability of DNN models for sparse datasets of multicomponent systems will extend the applicability of ML in the multicomponent material design. Furthermore, the general capability of MEIA to extend DNN models to multicomponent systems will provide new opportunities to utilize the progress of actively developed single-component DNNs for the modeling of multicomponent systems.

Integration of Machine Learning and Coarse-Grained Molecular Simulations for Polymer Materials: Physical Understandings and Molecular Design

“…This strategy leads to a total vector that is invariant to the arrangement of monomer components in copolymers, which is only applicable to random polymers where no sequence order is involved ( Kuenneth et al, 2021 ). For maintaining such feature invariance for different component permutations in random copolymers, the weighted summation method can be replaced by DNN networks to do mixing and aggregating, and the standard fingerprint can be changed to embedding networks to do feature representation learning ( Hanaoka, 2020 ). For more applications where CG bead sequence affects the properties of copolymers (sequence-defined copolymers), explicit-sequence featurization strategies are preferred.…”

Machine learning strategies for the structure-property relationship of copolymers

“…They assumed all copolymers to be random copolymers; thus, only monomer composition information is included in ML models without considering their sequences. Based on random copolymers and homopolymers, Hanaoka (2020) , Leibfarth et al., ( Reis et al., 2021 ), Kosuri et al. (2022) , Shi et al.…”

Capturing Molecular Interactions in Graph Neural Networks: A Case Study in Multi-Component Phase Equilibrium

“…7,8 A fundamental step in the use of ML models is the pre-definition or precalculation of molecular descriptors; [9][10][11][12] such descriptors are used as input data to develop quantitative structure-property relationship models. 13 Recently, there has been growing interest in applying ML models to study more complex chemical systems that might contain multiple components such as chemical reactions, 14,15 alloys, 16,17 copolymers, [18][19][20] and gas/liquid mixtures. [21][22][23][24][25][26][27][28][29][30][31][32] Among the ML techniques explored, graph neural networks (GNNs) 33,34 have gained special popularity because they can directly incorporate molecular representations (in the form of graphs), which enable the capturing of key structural information while potentially avoiding the need to pre-calculate/pre-define descriptors using more advanced but computationally-intensive tools such density functional theory (DFT) or molecular dynamics (MD) models.…”

“…36,37 When dealing with multiple components, several approaches have been devised; a typical way to encode multi-molecule information is to simply average or concatenate the features of individual molecules and to use these as system-level features for property inference with fully connected or attentive layers. 14,15,19 Previous studies have also incorporated weighted sums or concatenation to take into account composition information when needed. 19 However, these approaches do not capture molecular interactions in an explicit manner, which may limit the predictive power of GNNs for systems in which intermolecular interactions play an important role.…”

“…14,15,19 Previous studies have also incorporated weighted sums or concatenation to take into account composition information when needed. 19 However, these approaches do not capture molecular interactions in an explicit manner, which may limit the predictive power of GNNs for systems in which intermolecular interactions play an important role.…”