Message-passing neural networks for high-throughput polymer screening

Abstract

Machine learning methods have shown promise in predicting molecular properties, and given sufficient training data, machine learning approaches can enable rapid high-throughput virtual screening of large libraries of compounds. Graph-based neural network architectures have emerged in recent years as the most successful approach for predictions based on molecular structure and have consistently achieved the best performance on benchmark quantum chemical datasets. However, these models have typically required optimized 3D structural information for the molecule to achieve the highest accuracy. These 3D geometries are costly to compute for high levels of theory, limiting the applicability and practicality of machine learning methods in high-throughput screening applications. In this study, we present a new database of candidate molecules for organic photovoltaic applications, comprising approximately 91 000 unique chemical structures. Compared to existing datasets, this dataset contains substantially larger molecules (up to 200 atoms) as well as extrapolated properties for long polymer chains. We show that message-passing neural networks trained with and without 3D structural information for these molecules achieve similar accuracy, comparable to state-of-the-art methods on existing benchmark datasets. These results therefore emphasize that for larger molecules with practical applications, near-optimal prediction results can be obtained without using optimized 3D geometry as an input. We further show that learned molecular representations can be leveraged to reduce the training data required to transfer predictions to a new density functional theory functional.

Prediction of organic homolytic bond dissociation enthalpies at near chemical accuracy with sub-second computational cost

“…For enthalpy prediction on the QM9 dataset, consisting of all small molecules satisfying known valence rules, 3D coordinates appear to lead to superior prediction performance 20 . However, a recent study has shown that for some molecules and properties, 3D coordinates did not necessarily lead to improved results over more simple representations of 2D connectivity and atom types (i.e., SMILES 26 notation) 21 . In addition, while precise, absolute QM-derived atomization energies are often inaccurate by up to a full Hartree for common molecules (627 kcal mol −1 ) 27 .…”

“…The rise of machine learning (ML) in quantum chemistry has led to the development of highly-accurate empirical models 19 that have accelerated traditionally difficult QM calculations for predicting enthalpy 20 , optoelectronic properties 21 , and forces 22 . In particular, the rise of graph neural networks (GNNs) 23 in modeling chemical properties has enabled 'end-to-end' learning on molecular structure: a ML strategy where traditional feature engineering is replaced by feature learning from a graph-based molecular representation 19 .…”

Bioactivity descriptors for uncharacterized compounds

“…The renaissance of (deep) neural networks has fueled the development of novel structure 'featurizers' 28 based on graph/image convolutions of molecules [29][30][31] , the apprehension of the SMILES syntax 32 , or even a unified representation of protein targets 33 . These techniques are able to identify problem-specific patterns and, in general, they outperform conventional chemical fingerprints.…”

Aptamers for Detection and Diagnostics (ADD) is a proposed mobile app acquiring optical data from conjugated quantum nanodots to identify molecules indicating presence of SARS-CoV-2 virus: Why public health and healthcare need smartphone sensors as a platform for early detection and prevention

“…The learning that repurposed the antibiotic 556 Halicin, is one good example. Training the MPNN (message passing neural networks 557 ) was structured and deployed hyperparameter 558 optimization, without any artificial intelligence 559 , as explained elsewhere 560 . Ensembling 561 was applied to improve outcomes in silico but predictions were biologically tested.…”