ABSTRACT
Molecular representation learning is the first yet vital step in combining deep learning and molecular science. To push the boundaries of molecular representation learning, we present PhysChem, a novel neural architecture that learns molecular representations via fusing physical and chemical information of molecules. PhysChem is composed of a physicist network (PhysNet) and a chemist network (ChemNet). PhysNet is a neural physical engine that learns molecular conformations through simulating molecular dynamics with parameterized forces;ChemNet implements geometry-aware deep message-passing to learn chemical/biomedical properties of molecules. Two networks specialize in their own tasks and cooperate by providing expertise to each other. By fusing physical and chemical information, PhysChem achieved state-of-the-art performances on MoleculeNet, a standard molecular machine learning benchmark. The effectiveness of PhysChem was further corroborated on cutting-edge datasets of SARS-CoV-2. © 2021 Neural information processing systems foundation. All rights reserved.
ABSTRACT
Two eucalyptol derivatives, namely 1,3,3-trimethyl-2-oxabicycle[2.2.2]oct-5-yl acetate (4) and 1,3,3-trimethyl-2-oxabicycle[2.2.2]oct-5,8-yl acetate (6) have been synthesized and characterized. Their crystal structures have been solved by single-crystal X-ray diffraction methods indicating that the molecular conformation of both compounds is stabilized by intramolecular C–H⋯O bonds between the H-atoms of the methyl group from the eucalyptol moiety and the O-atom of the acetoxy group. In addition, we have performed a detailed investigation of the intermolecular interactions that stabilize the crystal packing of both structures by using Hirshfeld surface analysis and their associated two-dimensional fingerprint plots. The analysis reveals that the solid-state structure of compounds 4 and 6 is mainly stabilized by C–H⋯O hydrogen bonds and H⋯H bonding interactions. These interactions have also been described and studied energetically using DFT calculations. The nature and strength of these intermolecular contacts have been rationalized by using several computational tools including molecular electrostatic potential (MEP) surfaces, natural bond orbital analysis (NBO), Bader's theory of atoms in molecules (QTAIM) and non-covalent interaction plot (NCI plot) index methods. Furthermore, the intermolecular contacts observed in the crystal lattice of both compounds were experimentally studied through vibrational (IR and Raman) and 1H and 13C NMR spectra. The computational molecular docking analysis of the compounds has been carried out against five potential leishmanial drug targets and the main protease of SARS-CoV-2.
ABSTRACT
A fast way to reconstruct the three-dimensional molecular conformation of SARS-CoV-2 virus proteins is addressed in this article, involving the most worrying variant discovered in patients from Brazil, the lineage B.1.1.28/P.1. The proposed methodology is based on the sequencing of virus proteins and that, through the incorporation of mutations in silico, which are then computationally reconstructed using an enumerative feasibility algorithm validated by the Ramachandran diagram and structural alignment, in addition to the subsequent study of structural stability through classical molecular dynamics. From the resulting structure to the ACE2-RBD complex, the valid solution presented 97.06% of the residues in the most favorable region while the reference crystallographic structure presented 95.0%, a difference therefore very small and revealing the great consistency of the developed algorithm. Another important result was the low RMSD alignment between the best solution by the BP algorithm and the reference structure, where we obtained 0.483Å. Finally, the molecular dynamics indicated greater structural stability in the ACE2-RBD interaction with the P.1 strain, which could be a plausible explanation for convergent evolution that provides an increase in the interaction affinity with the ACE2 receptor. ©2021 IEEE