Dataset Information

Modelling local and general quantum mechanical properties with attention-based pooling.

ABSTRACT: Atom-centred neural networks represent the state-of-the-art for approximating the quantum chemical properties of molecules, such as internal energies. While the design of machine learning architectures that respect chemical principles has continued to advance, the final atom pooling operation that is necessary to convert from atomic to molecular representations in most models remains relatively undeveloped. The most common choices, sum and average pooling, compute molecular representations that are naturally a good fit for many physical properties, while satisfying properties such as permutation invariance which are desirable from a geometric deep learning perspective. However, there are growing concerns that such simplistic functions might have limited representational power, while also being suboptimal for physical properties that are highly localised or intensive. Based on recent advances in graph representation learning, we investigate the use of a learnable pooling function that leverages an attention mechanism to model interactions between atom representations. The proposed pooling operation is a drop-in replacement requiring no changes to any of the other architectural components. Using SchNet and DimeNet++ as starting models, we demonstrate consistent uplifts in performance compared to sum and mean pooling and a recent physics-aware pooling operation designed specifically for orbital energies, on several datasets, properties, and levels of theory, with up to 85% improvements depending on the specific task.

SUBMITTER: Buterez D

PROVIDER: S-EPMC10686994 | biostudies-literature | 2023 Nov

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Modelling local and general quantum mechanical properties with attention-based pooling.

Buterez David D Janet Jon Paul JP Kiddle Steven J SJ Oglic Dino D Liò Pietro P

Communications chemistry 20231129 1

Atom-centred neural networks represent the state-of-the-art for approximating the quantum chemical properties of molecules, such as internal energies. While the design of machine learning architectures that respect chemical principles has continued to advance, the final atom pooling operation that is necessary to convert from atomic to molecular representations in most models remains relatively undeveloped. The most common choices, sum and average pooling, compute molecular representations that ...[more]

PMID: 38030692

Similar Datasets

Project description:Fluorescent RNA aptamers have been successfully applied to track and tag RNA in a biological system. However, it is still challenging to predict the excited-state properties of the RNA aptamer-fluorophore complex with the traditional electronic structure methods due to expensive computational costs. In this study, an accurate and efficient fragmentation quantum mechanical (QM) approach of the electrostatically embedded generalized molecular fractionation with conjugate caps (EE-GMFCC) scheme was applied for calculations of excited-state properties of the RNA aptamer-fluorophore complex. In this method, the excited-state properties were first calculated with one-body fragment quantum mechanics/molecular mechanics (QM/MM) calculation (the excited-state properties of the fluorophore) and then corrected with a series of two-body fragment QM calculations for accounting for the QM effects from the RNA on the excited-state properties of the fluorophore. The performance of the EE-GMFCC on prediction of the absolute excitation energies, the corresponding transition electric dipole moment (TEDM), and atomic forces at both the TD-HF and TD-DFT levels was tested using the Mango-II RNA aptamer system as a model system. The results demonstrate that the calculated excited-state properties by EE-GMFCC are in excellent agreement with the traditional full-system time-dependent ab initio calculations. Moreover, the EE-GMFCC method is capable of providing an accurate prediction of the relative conformational excited-state energies for different configurations of the Mango-II RNA aptamer system extracted from the molecular dynamics (MD) simulations. The fragmentation method further provides a straightforward approach to decompose the excitation energy contribution per ribonucleotide around the fluorophore and then reveals the influence of the local chemical environment on the fluorophore. The applications of EE-GMFCC in calculations of excitation energies for other RNA aptamer-fluorophore complexes demonstrate that the EE-GMFCC method is a general approach for accurate and efficient calculations of excited-state properties of fluorescent RNAs.

Dataset Information

Modelling local and general quantum mechanical properties with attention-based pooling.

Publications

Modelling local and general quantum mechanical properties with attention-based pooling.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets