AI/ML Models to Predict the Severity of Drug-Induced Liver Injury for Small Molecules

Drug-induced liver injury (DILI), believed to be a multifactorial toxicity, has been a leading cause of attrition of small molecules during discovery, clinical development, and postmarketing. Identification of DILI risk early reduces the costs and cycle times associated with drug development. In recent years, several groups have reported predictive models that use physicochemical properties or in vitro and in vivo assay endpoints; however, these approaches have not accounted for liver-expressed proteins and drug molecules. To address this gap, we have developed an integrated artificial intelligence/machine learning (AI/ML) model to predict DILI severity for small molecules using a combination of physicochemical properties and off-target interactions predicted in silico. We compiled a data set of 603 diverse compounds from public databases. Among them, 164 were categorized as Most DILI (M-DILI), 245 as Less DILI (L-DILI), and 194 as No DILI (N-DILI) by the FDA. Six machine learning methods were used to create a consensus model for predicting the DILI potential. These methods include k-nearest neighbor (k-NN), support vector machine (SVM), random forest (RF), Naïve Bayes (NB), artificial neural network (ANN), logistic regression (LR), weighted average ensemble learning (WA) and penalized logistic regression (PLR). Among the analyzed ML methods, SVM, RF, LR, WA, and PLR identified M-DILI and N-DILI compounds, achieving a receiver operating characteristic area under the curve of 0.88, sensitivity of 0.73, and specificity of 0.9. Approximately 43 off-targets, along with physicochemical properties (fsp³, log S, basicity, reactive functional groups, and predicted metabolites), were identified as significant factors in distinguishing between M-DILI and N-DILI compounds. The key off-targets that we identified include: PTGS1, PTGS2, SLC22A12, PPARγ, RXRA, CYP2C9, AKR1C3, MGLL, RET, AR, and ABCC4. The present AI/ML computational approach therefore demonstrates that the integration of physicochemical properties and predicted on- and off-target biological interactions can significantly improve DILI predictivity compared to chemical properties alone.

1,3-Dipolar Cycloaddition of 3-Azido-3-deoxy-1,2:5,6-di-O-isopropylidene-α-D-glucofuranose and C60

The thermal cycloaddition of azido sugar 3 with C60 gives the chiral azafulleroid 4 in 2.55% yield (8% based in recovered C60).

Identification of N-(2-(azepan-1-yl)-2-phenylethyl)-benzenesulfonamides as novel inhibitors of GlyT1

A novel series of glycine transporter 1 (GlyT1) inhibitors is described. Scoping of the heterocycle moiety of hit 4-chlorobenzenesulfonamide 1 led to replacement of the piperidine with an azepane for a modest increase in potency. Phenyl sulfonamides proved superior to alkyl and non-phenyl aromatic sulfonamides, while subsequent ortho substitution of the 2-(azepan-1-yl)-2-phenylethanamine aromatic ring yielded 39 (IC(50) 37 nM, solubility 14 microM), the most potent GlyT1 inhibitor in this series. Favorable brain-plasma ratios were observed for select compounds in pharmacokinetic studies to evaluate CNS penetration.

Hydride transfer catalyzed by xylose isomerase: mechanism and quantum effects

We have applied molecular dynamics umbrella-sampling simulation and ensemble-averaged variational transition state theory with multidimensional tunneling (EA-VTST/MT) to calculate the reaction rate of xylose-to- xylulose isomerization catalyzed by xylose isomerase in the presence of two Mg2+ ions. The calculations include determination of the free energy of activation profile and ensemble averaging in the transmission coefficient. The potential energy function is approximated by a combined QM/MM/SVB method involving PM3 for the quantum mechanical (QM) subsystem, CHARMM22 and TIP3P for the molecular mechanical (MM) environment, and a simple valence bond (SVB) local function of two bond distances for the hydride transfer reaction. The simulation confirms the essential features of a mechanism postulated on the basis of kinetics and X-ray data by Whitlow et al. (Whitlow, M.; Howard, A. J.; Finzel, B. C.; Poulos, T. L.; Winborne, E.; Gilliland, G. L. Proteins 1991, 9, 153) and Ringe, Petsko, and coworkers (Labie, A.; Allen, K.-N.; Petsko, G. A.; Ringe, D. Biochemistry 1994, 33, 5469). This mechanism involves a rate-determining 1,2-hydride shift with prior and post proton transfers. Inclusion of quantum mechanical vibrational energy is important for computing the free energy of activation, and quantum mechanical tunneling effects are essential for computing kinetic isotope effects (KIEs). It is found that 85% of the reaction proceeds by tunneling and 15% by overbarrier events. The computed KIE for the ratio of hydride to deuteride transfer is in good agreement with the experimental results. The molecular dynamics simulations reveal that proton and hydride transfer reactions are assisted by breathing motions of the mobile Mg2+ ion in the active site, providing evidence for concerted motion of Mg2+ during the hydride transfer step.

Quantum dynamics of hydride transfer catalyzed by bimetallic electrophilic catalysis: synchronous motion of Mg(2+) and H(-) in xylose isomerase

Xylose isomerase exhibits a bridged-bimetallic active-site motif in which the substrate is bound to two metals connected by a glutamate bridge, and X-ray crystallographic studies suggest that metal movement is involved in the hydride transfer rate-controlling catalytic step. Here we report classical/quantal dynamical simulations of this step that provide new insight into the metal motion. The potential energy surface is calculated by treating xylose with semiempirical molecular orbital theory augmented by a simple valence bond potential and the rest of the system by molecular mechanics. The rate constant for the hydride-transfer step was calculated by ensemble-averaged dynamical simulations including both variational transition-state theory for determination of the statistically averaged dynamical bottleneck and optimized multidimensional tunneling calculations. The dynamics calculations include 25 317 atoms, with quantized vibrational free energy in 89 active-site degrees of freedom, and with 32 atoms moving through static secondary zone transition-state configurations in the quantum tunneling simulation. Our simulations show that the average Mg-Mg distance R increases monotonically as a function of the hydride-transfer progress variable z. The range of the average R along the reaction path is consistent with the X-ray structure, thus providing a dynamical demonstration of the postulated role of Mg in catalysis. We also predicted the primary deuterium kinetic isotope effect (KIE) for the chemical step. We calculated a KIE of 3.8 for xylose at 298 K, which is consistent with somewhat smaller experimentally observed KIEs for glucose substrate at higher temperatures. More than half of our KIE is due to tunneling; neglecting quantum effects on the reaction coordinate reduces the calculated KIE to 1.8.

Role of sugar re-puckering in the transition of A and B forms of DNA in solution. A molecular dynamics study

State of the art molecular dynamic simulations show that simple modification of the sugar puckering of 2'deoxyriboses leads to a reversible change between two stable forms of DNA which resemble very closely the canonical A and B duplex forms. Analysis of the A-type and B-type structures reveals interesting, and previously unknown features of these two families of conformations of the DNA.

Molecular dynamics study of the binding of elsamicin A to DNA

Molecular mechanics (MM) and dynamics (MD) were used to examine the conformational flexibility of elsamicin A, an antitumour antibiotic, in aqueous solution. Furthermore, MM and MD simulations were performed to obtain the first three-dimensional model of the elsamicin-A-DNA complex. Comparison of MD simulations of free and DNA-bound elsamicin A provided insight into the mechanism of DNA cleavage. Moreover, the drug-DNA complexes were analyzed to investigate specific drug-DNA interactions, as well as the role of the solvent water molecules in the drug-DNA binding. On the basis of the results derived from these simulations, several issues related to the sequence-selective binding of elsamicin A to DNA are discussed.

Effects of different continium dielectric models in a molecular dynamics and energy minimization study of the antigenic loop of foot-and-mouth disease virus

This study presents the influence of the dielectric constant on the final structure of the major antigenic loop of the FMDV serotype C. Minimizations have been performed on the nine-residue peptide Ac-TASARGDLA-NHMe, using two expressions for the dielectric constant: a distance-independent (epsilon = kappa), and a distance-dependent (epsilon = kappa *r) forms, and for kappa values from 1 to 10. In addition, kappa = 40 and 80 has also been considered for the constant expression of the dielectrics. The calculations suggest, for both expressions of the dielectrics, that for the sequence under study a value of kappa in the range of 2-6 performs an adequate treatment of the electrostatic interactions. Finally, molecular dynamic simulations at 298 K calculated with an effective dielectric constants of epsilon = 1*r and epsilon = 4*r are compared. The results indicated that a dielectric of epsilon = 4*r provides a peptide fluctuations which are in agreement with recent X-ray diffraction studies.