No abstract

Machine learning continues to make strident advances in the prediction of desired properties concerning drug development. Problematically, the efficacy of machine learning in these arenas is reliant upon highly accurate and abundant data. These two limitations, high accuracy and abundance, are often taken together; however, insight into the dataset accuracy limitation of contemporary machine learning algorithms may yield insight into whether non-bench experimental sources of data may be used to generate useful machine learning models where there is a paucity of experimental data. We took highly accurate data across six kinase types, one GPCR, one polymerase, a human protease, and HIV protease, and intentionally introduced error at varying population proportions in the datasets for each target. With the generated error in the data, we explored how the retrospective accuracy of a Naïve Bayes Network, a Random Forest Model, and a Probabilistic Neural Network model decayed as a function of error. Additionally, we explored the ability of a training dataset with an error profile resembling that produced by the Free Energy Perturbation method (FEP+) to generate machine learning models with useful retrospective capabilities. The categorical error tolerance was quite high for a Naïve Bayes Network algorithm averaging 39% error in the training set required to lose predictivity on the test set. Additionally, a Random Forest tolerated a significant degree of categorical error introduced into the training set with an average error of 29% required to lose predictivity. However, we found the Probabilistic Neural Network algorithm did not tolerate as much categorical error requiring an average of 20% error to lose predictivity. Finally, we found that a Naïve Bayes Network and a Random Forest could both use datasets with an error profile resembling that of FEP+. This work demonstrates that computational methods of known error distribution like FEP+ may be useful in generating machine learning models not based on extensive and expensive in vitro-generated datasets.

N-Methyl-D-aspartate receptors (NMDARs) are ionotropic ligand-gated glutamate receptors that mediate fast excitatory synaptic transmission in the central nervous system (CNS). Several neurological disorders may involve NMDAR hypofunction, which has driven therapeutic interest in positive allosteric modulators (PAMs) of NMDAR function. Here we describe modest changes to the tetrahydroisoquinoline scaffold of GluN2C/ GluN2D-selective PAMs that expands activity to include GluN2A-and GluN2B-containing recombinant and synaptic NMDARs. These new analogues are distinct from GluN2C/GluN2D-selective compounds like (+)-(3chlorophenyl)(6,7-dimethoxy-1-((4-methoxyphenoxy)methyl)-3,4-dihydroisoquinolin-2(1H)-yl)methanone (CIQ) by virtue of their subunit selectivity, molecular determinants of action, and allosteric regulation of agonist potency. The (S)-enantiomers of two analogues (EU1180-55, EU1180-154) showed activity at NMDARs containing all subunits (GluN2A, GluN2B, GluN2C, GluN2D), whereas the (R)enantiomers were primarily active at GluN2C-and GluN2D-containing NMDARs. Determination of the actions of enantiomers on triheteromeric receptors confirms their unique pharmacology, with greater activity of (S) enantiomers at GluN2A/GluN2D and GluN2B/GluN2D subunit combinations than (R) enantiomers. Evaluation of the (S)-EU1180-55 and EU1180-154 response of chimeric kainate/NMDA receptors revealed structural determinants of action within the pore-forming region and associated linkers. Scanning mutagenesis identified structural determinants within the GluN1 pre-M1 and M1 regions that alter the activity of (S)-EU1180-55 but not (R)-EU1180-55. By contrast, mutations in pre-M1 and M1 regions of GluN2D perturb the actions of only the (R)-EU1180-55 but not the (S) enantiomer. Molecular modeling supports the idea that the (S) and (R) enantiomers interact distinctly with GluN1 and GluN2 pre-M1 regions, suggesting that two distinct sites exist for these NMDAR PAMs, each of which has different functional effects.

-methyl-d-aspartate (NMDA) receptors are ligand-gated, cation-selective channels that mediate a slow component of excitatory synaptic transmission. Subunit-selective positive allosteric modulators of NMDA receptor function have therapeutically relevant effects on multiple processes in the brain. A series of pyrrolidinones, such as PYD-106, that selectively potentiate NMDA receptors that contain the GluN2C subunit have structural determinants of activity that reside between the GluN2C amino terminal domain and the GluN2C agonist binding domain, suggesting a unique site of action. Here we use molecular biology and homology modeling to identify residues that line a candidate binding pocket for GluN2C-selective pyrrolidinones. We also show that occupancy of only one site in diheteromeric receptors is required for potentiation. Both GluN2A and GluN2B can dominate the sensitivity of triheteromeric receptors to eliminate the actions of pyrrolidinones, thus rendering this series uniquely sensitive to subunit stoichiometry. We experimentally identified NMR-derived conformers in solution, which combined with molecular modeling allows the prediction of the bioactive binding pose for this series of GluN2C-selective positive allosteric modulators of NMDA receptors. These data advance our understanding of the site and nature of the ligand-protein interaction for GluN2C-selective positive allosteric modulators for NMDA receptors.

A method capable of identifying novel synthetic targets for small molecule lead optimization has been developed. The FRESH (FRagment-based Exploitation of modular Synthesis by vHTS) approach relies on a multistep synthetic route to a target series of compounds devised by a close collaboration between synthetic and computational chemists. It combines compound library generation, quantitative structure−acitvity relationship construction, fragment processing, virtual high throughput screening and display of results within the Pipeline Pilot framework. Outcomes enumerate tailored selection of novel synthetic targets with improved potency and optimized physical properties for an emerging compound series. To validate the application of FRESH, three retrospective case studies have been performed to pinpoint reported potent analogues. One prospective case study was performed to demonstrate that FRESH is able to capture additional potent analogues.