Predicting 2H NMR acyl chain order parameters with graph neural networks. (October 2022)

• Record Type: Journal Article
• Authors: Fischer, Markus; Schwarze, Benedikt; Ristic, Nikola; Scheidt, Holger A.
• Is Part Of: Computational biology and chemistry. Volume 100(2022)
• Publication Date: 2022-10
• Subjects: G2R graph-to-order neural network architecture -- GNN graph neural network -- MLP multi-layer perceptron -- NMR nuclear magnetic resonance -- POPC-d31 1-palmitoyl-d31-2-oleoyl-sn-glycero-3-phosphocholine; Graph neural network -- Order parameters -- Deuterium NMR -- Prediction; Chemistry -- Data processing -- Periodicals; Biology -- Data processing -- Periodicals; Biochemistry -- Data processing; Biology -- Data processing; Molecular biology -- Data processing; Periodicals; Electronic journals; 542.85
• DOI: 10.1016/j.compbiolchem.2022.107750
• Languages: English
• ISSNs: 1476-9271

Abstract

Abstract: 2 H NMR order parameters of the acyl chain of phospholipid membranes are an important indicator of the effects of molecules on membrane order, mobility, and permeability. So far, the evaluation procedures are case-by-case studies for every type of small molecule with certain types of membranes. Rapid screening of the effects of a variety of drugs would be invaluable if it were possible. Unfortunately, to date there is no practical or theoretical approach to this as there is with other experimental parameters, e.g., chemical shifts from 1 H and 13 C NMR. We aim to remedy this situation by introducing a model based on graph neural networks (GNN) capable of predicting 2 H NMR order parameters of lipid membranes in the presence of different molecules based on learned molecular features. Rapid prediction of these parameters would allow fast assessment of potential effects of drugs on lipid membranes, which is important for further drug development and provides insight into potential side effects. We conclude that the graph network-based model presented in this work can predict order parameters with sufficient accuracy, and we are confident that the concepts presented are a suitable basis for future research. We also make our model available to the public as a web application at https://proteinformatics.uni-leipzig.de/g2r/ . Graphical Abstract: ga1 Highlights: 2 H NMR chain order are an important tool to characterize the membrane interaction. molecular features can be extracted by graph neural networks. machine learning on molecular features can be used to predict membrane impact. graph neural networks can predict 2 H NMR order parameters of lipid membranes.