Chemprop: A Machine Learning Package for Chemical Property Prediction.

Deep learning has become a powerful and frequently employed tool for the prediction of molecular properties, thus creating a need for open-source and versatile software solutions that can be operated by nonexperts. Among the current approaches, directed message-passing neural networks (D-MPNNs) have proven to perform well on a variety of property prediction tasks. The software package Chemprop implements the D-MPNN architecture and offers simple, easy, and fast access to machine-learned molecular properties. Compared to its initial version, we present a multitude of new Chemprop functionalities such as the support of multimolecule properties, reactions, atom/bond-level properties, and spectra. Further, we incorporate various uncertainty quantification and calibration methods along with related metrics as well as pretraining and transfer learning workflows, improved hyperparameter optimization, and other customization options concerning loss functions or atom/bond features. We benchmark D-MPNN models trained using Chemprop with the new reaction, atom-level, and spectra functionality on a variety of property prediction data sets, including MoleculeNet and SAMPL, and observe state-of-the-art performance on the prediction of water-octanol partition coefficients, reaction barrier heights, atomic partial charges, and absorption spectra. Chemprop enables out-of-the-box training of D-MPNN models for a variety of problem settings in fast, user-friendly, and open-source software.

Autonomous, multiproperty-driven molecular discovery: From predictions to measurements and back.

A closed-loop, autonomous molecular discovery platform driven by integrated machine learning tools was developed to accelerate the design of molecules with desired properties. We demonstrated two case studies on dye-like molecules, targeting absorption wavelength, lipophilicity, and photooxidative stability. In the first study, the platform experimentally realized 294 unreported molecules across three automatic iterations of molecular design-make-test-analyze cycles while exploring the structure-function space of four rarely reported scaffolds. In each iteration, the property prediction models that guided exploration learned the structure-property space of diverse scaffold derivatives, which were realized with multistep syntheses and a variety of reactions. The second study exploited property models trained on the explored chemical space and previously reported molecules to discover nine top-performing molecules within a lightly explored structure-property space.

Structure-Activity Relationship and Biological Investigation of a REV-ERBα-Selective Agonist SR-29065 (34) for Autoimmune Disorders.

Autoimmune diseases affect 50 million Americans, predominantly women, and are thought to be one of the top 10 leading causes of death among women in age groups up to 65 years. A central role for TH17 cells has been highlighted by genome-wide association studies (GWAS) linking genes preferentially expressed in TH17 cells to several human autoimmune diseases. We and others have reported that the nuclear receptors REV-ERBα and ß are cell-intrinsic repressors of TH17 cell development and pathogenicity and might therefore be therapeutic targets for intervention. Herein, we describe detailed SAR studies of a novel REV-ERBα-selective scaffold. Metabolic stability of the ligands was optimized allowing for in vivo interrogation of the receptor in a mouse model of multiple sclerosis (EAE) with a ligand (34). Reduction in frequency and number of T-cells in the CNS as well as key REV-ERB target genes is a measure of target engagement in vivo.

Multi-fidelity prediction of molecular optical peaks with deep learning.

Optical properties are central to molecular design for many applications, including solar cells and biomedical imaging. A variety of ab initio and statistical methods have been developed for their prediction, each with a trade-off between accuracy, generality, and cost. Existing theoretical methods such as time-dependent density functional theory (TD-DFT) are generalizable across chemical space because of their robust physics-based foundations but still exhibit random and systematic errors with respect to experiment despite their high computational cost. Statistical methods can achieve high accuracy at a lower cost, but data sparsity and unoptimized molecule and solvent representations often limit their ability to generalize. Here, we utilize directed message passing neural networks (D-MPNNs) to represent both dye molecules and solvents for predictions of molecular absorption peaks in solution. Additionally, we demonstrate a multi-fidelity approach based on an auxiliary model trained on over 28 000 TD-DFT calculations that further improves accuracy and generalizability, as shown through rigorous splitting strategies. Combining several openly-available experimental datasets, we benchmark these methods against a state-of-the-art regression tree algorithm and compare the D-MPNN solvent representation to several alternatives. Finally, we explore the interpretability of the learned representations using dimensionality reduction and evaluate the use of ensemble variance as an estimator of the epistemic uncertainty in our predictions of molecular peak absorption in solution. The prediction methods proposed herein can be integrated with active learning, generative modeling, and experimental workflows to enable the more rapid design of molecules with targeted optical properties.

1-(4-Phenylpiperazin-1-yl)-2-(1H-pyrazol-1-yl)ethanones as novel CCR1 antagonists.

A novel series of CCR1 antagonists based on the 1-(4-phenylpiperazin-1-yl)-2-(1H-pyrazol-1-yl)ethanone scaffold was identified by screening a compound library utilizing CCR1-expressing human THP-1 cells. SAR studies led to the discovery of the highly potent and selective CCR1 antagonist 14 (CCR1 binding IC(50)=4 nM using [(125)I]-CCL3 as the chemokine ligand). Compound 14 displayed promising pharmacokinetic and toxicological profiles in preclinical species.

Capillary electrophoretic separation-based approach to determine the labeling kinetics of oligodeoxynucleotides.

With the recent advances in electron microscopy (EM), computation, and nanofabrication, the original idea of reading DNA sequence directly from an image can now be tested. One approach is to develop heavy atom labels that can provide the contrast required for EM imaging. While evaluating tentative labels for the respective nucleobases in synthetic oligodeoxynucleotides (oligos), we developed a streamlined CE protocol to assess the label stability, reactivity, and selectivity. We report our protocol using osmium tetroxide 2,2'-bipyridine (Osbipy) as a thymidine (T) specific label. The observed rates show that the labeling process is kinetically independent of both the oligo length, and the base composition. The conditions, i.e. temperature, optimal Osbipy concentration, and molar ratio of reagents, to promote 100% conversion of the starting oligo to labeled product were established. Hence, the optimized conditions developed with the oligos could be leveraged to allow osmylation of effectively all Ts in ssDNA, while achieving minimal mislabeling. In addition, the approach and methods employed here may be adapted to the evaluation of other prospective contrasting agents/labels to facilitate next-generation DNA sequencing by EM.

Binding of madindoline A to the extracellular domain of gp130.

Elevated levels of IL-6 and IL-11 are associated with multiple myeloma, rheumatoid arthritis, hypercalcemia, cancer cachexia, and Castleman's disease. Madindoline A (MadA), isolated from Streptomyces nitrosporeus K93-0711, specifically inhibits the growth of IL-6- and IL-11-dependent cell lines, most likely by interfering with the homodimerization of gp130. This raises the possibility that MadA can be used as a model compound for the development of novel chemotherapeutic agents. In this report, we demonstrate that the binding of MadA to gp130 is specific and noncovalent, and displays a relatively low affinity. Furthermore, we show that the tricyclic 3a-hydroxytetrahydrofuro[2,3-b]indole (HFI) moiety of MadA alone is not sufficient for binding. Matrix-bound MadA precipitates a protein composed of the extracellular domain of gp130 fused to the Fc region of the immunoglobulin heavy chain. Binding is inhibited in a dose-dependent manner by preincubation with free MadA. The K(D) for binding of MadA to gp130 is 288 microM, as determined by surface plasmon resonance (SPR)-based biosensor analysis. The HFI portion of MadA does not bind to gp130 in either affinity precipitation or SPR analyses. Finally, MadA, but not the HFI portion, inhibits IL-6-dependent Stat3 tyrosine phosphorylation in HepG2 cells.

Synthesis of phakellistatin 13 and oxidation to phakellistatin 3 and isophakellistatin 3.

The natural product phakellistatin 13 cyclo-(TrpProPheGlyProThrLeu) was synthesized. Photosensitized oxidation of phakellistatin 13 gave the natural products phakellistatin 3 and isophakellistatin 3, demonstrating for the first time that a tryptophan residue can be directly converted to the corresponding 3a-hydroxypyrrolo[2,3-b]indoline in a full length peptide. Competitive oxidation of the indoline product was identified as the cause of low mass balance and is probably the source of low mass balance in the oxidative cyclization of all tryptamine derivatives. [reaction--see text]

Iron(II)-catalyzed sulfimidation and [2,3]-sigmatropic rearrangement of propargyl sulfides with tert-butoxycarbonyl azide. Access to N-allenylsulfenimides.

The iron(II)-catalyzed Bach reaction of tert-butoxycarbonyl azide (BocN(3)) and allyl sulfides has been extended to include propargyl sulfides, which give N-allenylsulfenimide products. Using 10 mol % dppeFeCl(2) as catalyst the reaction proceeds at 0 degrees C with a number of different propargyl sulfides in 31-73% isolated yield. The reaction is limited by product instability toward catalyst and termination of the catalytic cycle by excess BocN(3). N-Allenylsulfenimide 2b smoothly undergoes catalytic hydrogenation and a Diels-Alder reaction with cyclopentadiene.

Atom economy. Palladium-catalyzed formation of coumarins by addition of phenols and alkynoates via a net C-H insertion.

A strategy to achieve ortho substitution of phenols initiated by an ortho-palladation to create coumarins was examined. Indeed, treatment of alkynoates with electron-rich phenols in the presence of a palladium catalyst and an acid does generate coumarins. The scope of the reaction with respect to the phenol and the alkynoates is defined. With unsymmetrical aromatic substrates, generally good regioselectivity that reflects the HOMO coefficients can be observed. In the course of these studies, numerous important naturally occurring coumarins have been synthesized, including fraxinol methyl ether, ayapin, herniarin, xanthoxyletin, and alloxanthoxyletin. The fact that a Pd(0) is the precatalyst rather than a Pd(+2) species and that an acid that reduces Pd(+2) salts, formic acid, functions better than other carboxylic acids raises doubts about the initial working hypothesis. A novel mechanism involving a palladium phenoxide formed from a hydridopalladium carboxylate and phenol is invoked to rationalize the results.