SchNetPack Hyperparameter Optimization for a More Reliable Top Docking Scores Prediction

Options to improve the extrapolation power of the neural network designed using the SchNetPack package with respect to top docking scores prediction are presented. It is shown that hyperparameter tuning of the atomistic model representation (in the schnetpack.representation) improves the prediction of the top scoring compounds, which have characteristically a low incidence in randomized data sets for training of machine learning models. The prediction robustness is evaluated according to the mean square error (MSE) and the entropy of the average loss landscape decrease. Admittedly, the improvement of the top scoring compounds' prediction accuracy comes with the penalty of worsening the overall prediction power. It is revealed that the most impactful hyperparameter is the cutoff (5 Å is reported as the optimal choice). Other parameters (e.g., number of radial basis functions, number of interaction layers of the neural network, feature vector size or its batch size) are found to not affect the prediction robustness of the top scoring compounds in any comparable way relative to the cutoff. The MSE of the best docking score prediction (below -13 kcal/mol) improves from ca. 3.5 to 0.9 kcal/mol, while the prediction of less potent compounds (-13 to -11 kcal/mol) shows a lesser improvement, i.e., a decrease of MSE from 1.6 to 1.3 kcal/mol. Additionally, oversampling and undersampling of the training set with respect to the top scoring compounds' abundance is presented. The results indicate that the cutoff choice performs better than over- or undersampling of the training set, with undersampling performing better than oversampling.

Chemical and Redox Noninnocence of Pentane-2,4-dione Bis(S-methylisothiosemicarbazone) in Cobalt Complexes and Their Application in Wacker-Type Oxidation

Cobalt complexes with multiproton- and multielectron-responsive ligands are of interest for challenging catalytic transformations. The chemical and redox noninnocence of pentane-2,4-dione bis(S-methylisothiosemicarbazone) (PBIT) in a series of cobalt complexes has been studied by a range of methods, including spectroscopy [UV-vis, NMR, electron paramagnetic resonance (EPR), X-ray absorption spectroscopy (XAS)], cyclic voltammetry, X-ray diffraction, and density functional theory (DFT) calculations. Two complexes [CoIII(H2LSMe)I]I and [CoIII(LSMe)I2] were found to act as precatalysts in a Wacker-type oxidation of olefins using phenylsilane, the role of which was elucidated through isotopic labeling. Insights into the mechanism of the catalytic transformation as well as the substrate scope of this selective reaction are described, and the essential role of phenylsilane and the noninnocence of PBIT are disclosed. Among the several relevant species characterized was an unprecedented Co(III) complex with a dianionic diradical PBIT ligand ([CoIII(LSMe••)I]).

Isomerization pathway of a C-C sigma bond in a bis(octaazamacrocycle)dinickel(II) complex activated by deprotonation: a DFT study

The anti (a) to syn (s) isomerization pathway of the deprotonated form of the dimer with two nickel(II) 15-membered octaazamacrocyclic units connected via a carbon-carbon (C-C) σ bond was investigated. For the initial anti (a) structure, a deprotonation of one of the bridging (sp3 hybridized) carbon atoms is suggested to allow for an a to s geometry twist. A 360° scan around the bridging C-C dihedral angle was performed first to find an intermediate geometry. Subsequently, the isomerization pathway was explored via individual steps using a series of mode redundant geometry optimizations (internal coordinates potential energy surface scans) and geometry relaxations leading to the s structure. The prominent geometries (intermediates) of the isomerization pathway are chosen and compared to the a and s structures, and geometry relaxations of the protonated forms of selected intermediates are considered.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00214-024-03100-5.

Compromise in Docking Power of Liganded Crystal Structures of Mpro SARS-CoV-2 Surpasses 90% Success Rate

Herein, we present the capacity of three different molecular docking programs (AutoDock, AutoDock Vina, and PLANTS) to identify and reproduce the binding modes of ligands present in 247 covalent and 169 noncovalent complex crystal structures of the severe acute respiratory syndrome coronavirus 2 main protease (Mpro). The compromise in docking power is evaluated with respect to their ability to generate poses similar to the crystal structure binding mode (heavy atoms' root-mean-square deviation < 2 Å) and their ability to recognize the native binding mode with an included compensation for the scoring function error. Noncovalently bound inhibitors are best modeled by AutoDock Vina (90.6% success rate in the active site), while the most relevant results for covalently bound inhibitors are produced by PLANTS (93.0%). AutoDock shows acceptable performance for both types of ligands, 81.1 and 76.4% for noncovalent and covalent complexes, respectively. All three programs manifest worse performance when reproducing surface-bound ligands. Comparison with other works illustrates the importance of crystal structure processing (12% of noncovalent and 26% of covalent ligands had to be manually corrected), proper sampling protocol settings, and inclusion of root-mean-square deviation (RMSD)/scoring function error compensations in crystal structure pose identification. Results are analyzed with respect to a clustering scheme of the noncovalently bound ligands and the chemical reaction type of the covalent ligand bound to the Cys145 residue. A comparison of screening power based on the docking scores of noncovalent ligands from the crystal structures with a "Directory of Useful Decoys, Enhanced" set of known decoys (6562 compounds) and ZINC15 in vivo subset (60,394 compounds) is provided. Ligand and protein input files are provided for future benchmarking purposes.

Machine learning prediction of 3CLpro SARS-CoV-2 docking scores

Molecular docking results of two training sets containing 866 and 8,696 compounds were used to train three different machine learning (ML) approaches. Neural network approaches according to Keras and TensorFlow libraries and the gradient boosted decision trees approach of XGBoost were used with DScribe’s Smooth Overlap of Atomic Positions molecular descriptors. In addition, neural networks using the SchNetPack library and descriptors were used. The ML performance was tested on three different sets, including compounds for future organic synthesis. The final evaluation of the ML predicted docking scores was based on the ZINC in vivo set, from which 1,200 compounds were randomly selected with respect to their size. The results obtained showed a consistent ML prediction capability of docking scores, and even though compounds with more than 60 atoms were found slightly overestimated they remain valid for a subsequent evaluation of their drug repurposing suitability.

Charge density of 4-methyl-3-[(tetrahydro-2H-pyran-2-yl)oxy]thiazole-2(3H)-thione. A comprehensive multipole refinement, maximum entropy method and density functional theory study

The structure of 4-methyl-3-[(tetrahydro-2H-pyran-2-yl)oxy]thiazole-2(3H)-thione (MTTOTHP) was investigated using X-ray diffraction and computational chemistry methods for determining properties of the nitrogen-oxygen bond, which is the least stable entity upon photochemical excitation. Experimentally measured structure factors have been used to determine and characterize charge density via the multipole model (MM) and the maximum entropy method (MEM). Theoretical investigation of the electron density and the electronic structure has been performed in the finite basis set density functional theory (DFT) framework. Quantum Theory of Atoms In Molecules (QTAIM), deformation densities and Laplacians maps have been used to compare theoretical and experimental results. MM experimental results and predictions from theory differ with respect to the sign and/or magnitude of the Laplacian at the N-O bond critical point (BCP), depending on the treatment of n values of the MM radial functions. Such Laplacian differences in the N-O bond case are discussed with respect to a lack of flexibility in the MM radial functions also reported by Rykounov et al. [Acta Cryst. (2011), B67, 425-436]. BCP Hessian eigenvalues show qualitatively matching results between MM and DFT. In addition, the theoretical analysis used domain-averaged fermi holes (DAFH), natural bond orbital (NBO) analysis and localized (LOC) orbitals to characterize the N-O bond as a single σ bond with marginal π character. Hirshfeld atom refinement (HAR) has been employed to compare to the MM refinement results and/or neutron dataset C-H bond lengths and to crystal or single molecule geometry optimizations, including considerations of anisotropy of H atoms. Our findings help to understand properties of molecules like MTTOTHP as progenitors of free oxygen radicals.

Dinuclear manganese(iii) complexes with bioinspired coordination and variable linkers showing weak exchange effects: a synthetic, structural, spectroscopic and computation study

High-resolution HFEPR indicates weak exchange interactions between Mn^III ions in agreement with DFT calculations.

cis-Tetrachlorido-bis(indazole)osmium(iv) and its osmium(iii) analogues: paving the way towards the cis-isomer of the ruthenium anticancer drugs KP1019 and/or NKP1339

The relationship between cis-trans isomerism and anticancer activity has been mainly addressed for square-planar metal complexes, in particular, for platinum(ii), e.g., cis- and trans-[PtCl2(NH3)2], and a number of related compounds, of which, however, only cis-counterparts are in clinical use today. For octahedral metal complexes, this effect of geometrical isomerism on anticancer activity has not been investigated systematically, mainly because the relevant isomers are still unavailable. An example of such an octahedral complex is trans-[RuCl4(Hind)2]-, which is in clinical trials now as its indazolium (KP1019) or sodium salt (NKP1339), but the corresponding cis-isomers remain inaccessible. We report the synthesis of Na[cis-OsIIICl4(κN2-1H-ind)2]·(Na[1]) suggesting a route to the cis-isomer of NKP1339. The procedure involves heating (H2ind)[OsIVCl5(κN1-2H-ind)] in a high boiling point organic solvent resulting in an Anderson rearrangement with the formation of cis-[OsIVCl4(κN2-1H-ind)2] ([1]) in high yield. The transformation is accompanied by an indazole coordination mode switch from κN1 to κN2 and stabilization of the 1H-indazole tautomer. Fully reversible spectroelectrochemical reduction of [1] in acetonitrile at 0.46 V vs. NHE is accompanied by a change in electronic absorption bands indicating the formation of cis-[OsIIICl4(κN2-1H-ind)2]- ([1]-). Chemical reduction of [1] in methanol with NaBH4 followed by addition of nBu4NCl afforded the osmium(iii) complex nBu4N[cis-OsIIICl4(κN2-1H-ind)2] (nBu4N[1]). A metathesis reaction of nBu4N[1] with an ion exchange resin led to the isolation of the water-soluble salt Na[1]. The X-ray diffraction crystal structure of [1]·Me2CO was determined and compared with that of trans-[OsIVCl4(κN2-1H-ind)2]·2Me2SO (2·2Me2SO), also prepared in this work. EPR spectroscopy was performed on the OsIII complexes and the results were analyzed by ligand-field and quantum chemical theories. We furthermore assayed effects of [1] and Na[1] on cell viability and proliferation in comparison with trans-[OsIVCl4(κN1-2H-ind)2] [3] and cisplatin and found a strong reduction of cell viability at concentrations between 30 and 300 μM in different cancer cell lines (HT29, H446, 4T1 and HEK293). HT-29 cells are less sensitive to cisplatin than 4T1 cells, but more sensitive to [1] and Na[1], as shown by decreased proliferation and viability as well as an increased late apoptotic/necrotic cell population.

Spectroscopic and Computational Studies of Spin States of Iron(IV) Nitrido and Imido Complexes

High-oxidation-state metal complexes with multiply bonded ligands are of great interest for both their reactivity as well as their fundamental bonding properties. This paper reports a combined spectroscopic and theoretical investigation into the effect of the apical multiply bonded ligand on the spin-state preferences of threefold symmetric iron(IV) complexes with tris(carbene) donor ligands. Specifically, singlet (S = 0) nitrido [{PhB(Im^R)₃}FeN], R = ^tBu (1), Mes (mesityl, 2) and the related triplet (S = 1) imido complexes, [{PhB(Im^R)₃}Fe(NR')]⁺, R = Mes, R' = 1-adamantyl (3), ^tBu (4), were investigated by electronic absorption and Mössbauer effect spectroscopies. For comparison, two other Fe(IV) nitrido complexes, [(TIMEN^Ar)FeN]⁺ (TIMEN^Ar = tris[2-(3-aryl-imidazol-2-ylidene)ethyl]amine; Ar = Xyl (xylyl), Mes), were investigated by ⁵⁷Fe Mössbauer spectroscopy, including applied-field measurements. The paramagnetic imido complexes 3 and 4 were also studied by magnetic susceptibility measurements (for 3) and paramagnetic resonance spectroscopy: high-frequency and -field electron paramagnetic resonance (for 3 and 4) and frequency-domain Fourier-transform (FD-FT) terahertz electron paramagnetic resonance (for 3), which reveal their zero-field splitting parameters. Experimentally correlated theoretical studies comprising ligand-field theory and quantum chemical theory, the latter including both density functional theory and ab initio methods, reveal the key role played by the Fe 3d_z² (a₁) orbital in these systems: the nature of its interaction with the nitrido or imido ligand dictates the spin-state preference of the complex. The ability to tune the spin state through the energy and nature of a single orbital has general relevance to the factors controlling spin states in complexes with applicability as single molecule devices.

A five-coordinate manganese(iii) complex of a salen type ligand with a positive axial anisotropy parameter D

Quantum chemical calculations reproduced well the electronic absorption spectrum and spin Hamiltonian parameters for MnL(NCS).

HFEPR and Computational Studies on the Electronic Structure of a High-Spin Oxidoiron(IV) Complex in Solution

Nonheme iron enzymes perform diverse and important functions in biochemistry. The active form of these enzymes comprises the ferryl, oxidoiron(IV), [FeO](2+) unit. In enzymes, this unit is in the high-spin, quintet, S = 2, ground state, while many synthetic model compounds exist in the spin triplet, S = 1, ground state. Recently, however, Que and co-workers reported an oxidoiron(IV) complex with a quintet ground state, [FeO(TMG3tren)](OTf)2, where TMG3tren = 1,1,1-tris{2-[N2-(1,1,3,3-tetramethylguanidino)]ethyl}amine and OTf = CF3SO3(-). The trigonal geometry imposed by this ligand, as opposed to the tetragonal geometry of earlier model complexes, favors the high-spin ground state. Although [FeO(TMG3tren)](2+) has been earlier probed by magnetic circular dichroism (MCD) and Mössbauer spectroscopies, the technique of high-frequency and -field electron paramagnetic resonance (HFEPR) is superior for describing the electronic structure of the iron(IV) center because of its ability to establish directly the spin-Hamiltonian parameters of high-spin metal centers with high precision. Herein we describe HFEPR studies on [FeO(TMG3tren)](OTf)2 generated in situ and confirm the S = 2 ground state with the following parameters: D = +4.940(5) cm(-1), E = 0.000(5), B4(0) = -14(1) × 10(-4) cm(-1), g⊥ = 2.006(2), and g∥ = 2.03(2). Extraction of a fourth-order spin-Hamiltonian parameter is unusual for HFEPR and impossible by other techniques. These experimental results are combined with state-of-the-art computational studies along with previous structural and spectroscopic results to provide a complete picture of the electronic structure of this biomimetic complex. Specifically, the calculations reproduce well the spin-Hamiltonian parameters of the complex, provide a satisfying geometrical picture of the S = 2 oxidoiron(IV) moiety, and demonstrate that the TMG3tren is an "innocent" ligand.