Geometric deep learning-guided Suzuki reaction conditions assessment for applications in medicinal chemistry.

Suzuki cross-coupling reactions are considered a valuable tool for constructing carbon-carbon bonds in small molecule drug discovery. However, the synthesis of chemical matter often represents a time-consuming and labour-intensive bottleneck. We demonstrate how machine learning methods trained on high-throughput experimentation (HTE) data can be leveraged to enable fast reaction condition selection for novel coupling partners. We show that the trained models support chemists in determining suitable catalyst-solvent-base combinations for individual transformations including an evaluation of the need for HTE screening. We introduce an algorithm for designing 96-well plates optimized towards reaction yields and discuss the model performance of zero- and few-shot machine learning. The best-performing machine learning model achieved a three-category classification accuracy of 76.3% (±0.2%) and an F 1-score for a binary classification of 79.1% (±0.9%). Validation on eight reactions revealed a receiver operating characteristic (ROC) curve (AUC) value of 0.82 (±0.07) for few-shot machine learning. On the other hand, zero-shot machine learning models achieved a mean ROC-AUC value of 0.63 (±0.16). This study positively advocates the application of few-shot machine learning-guided reaction condition selection for HTE campaigns in medicinal chemistry and highlights practical applications as well as challenges associated with zero-shot machine learning.

Best of both worlds: An expansion of the state of the art pKa model with data from three industrial partners.

In a unique collaboration between Simulations Plus and several industrial partners, we were able to develop a new version 11.0 of the previously published in silico pKa model, S+pKa, with considerably improved prediction accuracy. The model's training set was vastly expanded by large amounts of experimental data obtained from F. Hoffmann-La Roche AG, Genentech Inc., and the Crop Science division of Bayer AG. The previous v7.0 of S+pKa was trained on data from public sources and the Pharmaceutical division of Bayer AG. The model has shown dramatic improvements in predictive accuracy when externally validated on three new contributor compound sets. Less expected was v11.0's improvement in prediction on new compounds developed at Bayer Pharma after v7.0 was released (2013-2023), even without contributing additional data to v11.0. We illustrate chemical space coverage by chemistries encountered in the five domains, public and industrial, outline model construction, and discuss factors contributing to model's success.

Selective and brain-penetrant HCN1 inhibitors reveal links between synaptic integration, cortical function, and working memory.

Hyperpolarization-activated and cyclic-nucleotide-gated 1 (HCN1) ion channels are proposed to be critical for cognitive function through regulation of synaptic integration. However, resolving the precise role of HCN1 in neurophysiology and exploiting its therapeutic potential has been hampered by minimally selective antagonists with poor potency and limited in vivo efficiency. Using automated electrophysiology in a small-molecule library screen and chemical optimization, we identified a primary carboxamide series of potent and selective HCN1 inhibitors with a distinct mode of action. In cognition-relevant brain circuits, selective inhibition of native HCN1 produced on-target effects, including enhanced excitatory postsynaptic potential summation, while administration of a selective HCN1 inhibitor to rats recovered decrement working memory. Unlike prior non-selective HCN antagonists, selective HCN1 inhibition did not alter cardiac physiology in human atrial cardiomyocytes or in rats. Collectively, selective HCN1 inhibitors described herein unmask HCN1 as a potential target for the treatment of cognitive dysfunction in brain disorders.

A Machine Learning Framework to Improve Rat Clearance Predictions and Inform Physiologically Based Pharmacokinetic Modeling.

During drug discovery and development, achieving appropriate pharmacokinetics is key to establishment of the efficacy and safety of new drugs. Physiologically based pharmacokinetic (PBPK) models integrating in vitro-to-in vivo extrapolation have become an essential in silico tool to achieve this goal. In this context, the most important and probably most challenging pharmacokinetic parameter to estimate is the clearance. Recent work on high-throughput PBPK modeling during drug discovery has shown that a good estimate of the unbound intrinsic clearance (CLint,u,) is the key factor for useful PBPK application. In this work, three different machine learning-based strategies were explored to predict the rat CLint,u as the input into PBPK. Therefore, in vivo and in vitro data was collected for a total of 2639 proprietary compounds. The strategies were compared to the standard in vitro bottom-up approach. Using the well-stirred liver model to back-calculate in vivo CLint,u from in vivo rat clearance and then training a machine learning model on this CLint,u led to more accurate clearance predictions (absolute average fold error (AAFE) 3.1 in temporal cross-validation) than the bottom-up approach (AAFE 3.6-16, depending on the scaling method) and has the advantage that no experimental in vitro data is needed. However, building a machine learning model on the bias between the back-calculated in vivo CLint,u and the bottom-up scaled in vitro CLint,u also performed well. For example, using unbound hepatocyte scaling, adding the bias prediction improved the AAFE in the temporal cross-validation from 16 for bottom-up to 2.9 together with the bias prediction. Similarly, the log Pearson r2 improved from 0.1 to 0.29. Although it would still require in vitro measurement of CLint,u., using unbound scaling for the bottom-up approach, the need for correction of the fu,inc by fu,p data is circumvented. While the above-described ML models were built on all data points available per approach, it is discussed that evaluation comparison across all approaches could only be performed on a subset because ca. 75% of the molecules had missing or unquantifiable measurements of the fraction unbound in plasma or in vitro unbound intrinsic clearance, or they dropped out due to the blood-flow limitation assumed by the well-stirred model. Advantageously, by predicting CLint,u as the input into PBPK, existing workflows can be reused and the prediction of the in vivo clearance and other PK parameters can be improved.

A high quality, industrial data set for binding affinity prediction: performance comparison in different early drug discovery scenarios.

We release a new, high quality data set of 1162 PDE10A inhibitors with experimentally determined binding affinities together with 77 PDE10A X-ray co-crystal structures from a Roche legacy project. This data set is used to compare the performance of different 2D- and 3D-machine learning (ML) as well as empirical scoring functions for predicting binding affinities with high throughput. We simulate use cases that are relevant in the lead optimization phase of early drug discovery. ML methods perform well at interpolation, but poorly in extrapolation scenarios-which are most relevant to a real-world application. Moreover, we find that investing into the docking workflow for binding pose generation using multi-template docking is rewarded with an improved scoring performance. A combination of 2D-ML and 3D scoring using a modified piecewise linear potential shows best overall performance, combining information on the protein environment with learning from existing SAR data.

Benchmarking Accuracy and Generalizability of Four Graph Neural Networks Using Large In Vitro ADME Datasets from Different Chemical Spaces.

In this work, we benchmark a variety of single- and multi-task graph neural network (GNN) models against lower-bar and higher-bar traditional machine learning approaches employing human engineered molecular features. We consider four GNN variants - Graph Convolutional Network (GCN), Graph Attention Network (GAT), Message Passing Neural Network (MPNN), and Attentive Fingerprint (AttentiveFP). So far deep learning models have been primarily benchmarked using lower-bar traditional models solely based on fingerprints, while more realistic benchmarks employing fingerprints, whole-molecule descriptors and predictions from other related endpoints (e. g., LogD7.4) appear to be scarce for industrial ADME datasets. In addition to time-split test sets based on Genentech data, this study benefits from the availability of measurements from an external chemical space (Roche data). We identify GAT as a promising approach to implementing deep learning models. While all the deep learning models significantly outperform lower-bar benchmark traditional models solely based on fingerprints, only GATs seem to offer a small but consistent improvement over higher-bar benchmark traditional models. Finally, the accuracy of in vitro assays from different laboratories predicting the same experimental endpoints appears to be comparable with the accuracy of GAT single-task models, suggesting that most of the observed error from the models is a function of the experimental error propagation.

Discovery of RO7185876, a Highly Potent γ-Secretase Modulator (GSM) as a Potential Treatment for Alzheimer's Disease.

γ-Secretase (GS) is a key target for the potential treatment of Alzheimer's disease. While inhibiting GS led to serious side effects, its modulation holds a lot of potential to deliver a safe treatment. Herein, we report the discovery of a potent and selective gamma secretase modulator (GSM) (S)-3 (RO7185876), belonging to a novel chemical class, the triazolo-azepines. This compound demonstrates an excellent in vitro and in vivo DMPK profile. Furthermore, based on its in vivo efficacy in a pharmacodynamic mouse model and the outcome of the dose range finding (DRF) toxicological studies in two species, this compound was selected to undergo entry in human enabling studies (e.g., GLP toxicology and scale up activities).

Understanding Molecular Drivers of Melanin Binding To Support Rational Design of Small Molecule Ophthalmic Drugs.

Binding of drugs to ocular melanin is a prominent biological phenomenon that affects the local pharmacokinetics and pharmacodynamics in the eye. In this work, we report on the development of in vitro and in silico tools for an early assessment and prediction of melanin binding properties of small molecules. A robust high-throughput assay has been established to study the binding of large sets of compounds to melanin. The extremely randomized trees approach was used to develop an in silico model able to predict the extent of melanin binding from the molecular properties of the compounds. After the last iteration of the model, strong melanin binders could prospectively be identified with 91% accuracy. On the basis of in vitro data generated for approximately 3400 chemically diverse drug-like small molecules, pronounced correlations were observed between the extent of melanin binding and the basicity, lipophilicity, and aromaticity of the compounds.

Discovery of Risdiplam, a Selective Survival of Motor Neuron-2 ( SMN2) Gene Splicing Modifier for the Treatment of Spinal Muscular Atrophy (SMA).

SMA is an inherited disease that leads to loss of motor function and ambulation and a reduced life expectancy. We have been working to develop orally administrated, systemically distributed small molecules to increase levels of functional SMN protein. Compound 2 was the first SMN2 splicing modifier tested in clinical trials in healthy volunteers and SMA patients. It was safe and well tolerated and increased SMN protein levels up to 2-fold in patients. Nevertheless, its development was stopped as a precautionary measure because retinal toxicity was observed in cynomolgus monkeys after chronic daily oral dosing (39 weeks) at exposures in excess of those investigated in patients. Herein, we describe the discovery of 1 (risdiplam, RG7916, RO7034067) that focused on thorough pharmacology, DMPK and safety characterization and optimization. This compound is undergoing pivotal clinical trials and is a promising medicine for the treatment of patients in all ages and stages with SMA.

Exploring the Structural Space of the Galectin-1-Ligand Interaction.

Galectin-1 is a tumor-associated protein recognizing the Galß1-4GlcNAc motif of cell-surface glycoconjugates. Herein, we report the stepwise expansion of a multifunctional natural scaffold based on N-acetyllactosamine (LacNAc). We obtained a LacNAc mimetic equipped with an alkynyl function on the 3'-hydroxy group of the disaccharide facing towards a binding pocket adjacent to the carbohydrate-recognition domain. It served as an anchor motif for further expansion by the Sharpless-Huisgen-Meldal reaction, which resulted in ligands with a binding mode almost identical to that of the natural carbohydrate template. X-ray crystallography provided a structural understanding of the galectin-1-ligand interactions. The results of this study enable the development of bespoke ligands for members of the galectin target family.

Discovery of a Novel Class of Survival Motor Neuron 2 Splicing Modifiers for the Treatment of Spinal Muscular Atrophy.

Spinal muscular atrophy (SMA) is caused by mutation or deletion of the survival motor neuron 1 (SMN1) gene, resulting in low levels of functional SMN protein. We have reported recently the identification of small molecules (coumarins, iso-coumarins and pyrido-pyrimidinones) that modify the alternative splicing of SMN2, a paralogous gene to SMN1, restoring the survival motor neuron (SMN) protein level in mouse models of SMA. Herein, we report our efforts to identify a novel chemotype as one strategy to potentially circumvent safety concerns from earlier derivatives such as in vitro phototoxicity and in vitro mutagenicity associated with compounds 1 and 2 or the in vivo retinal findings observed in a long-term chronic tox study with 3 at high exposures only. Optimized representative compounds modify the alternative splicing of SMN2, increase the production of full length SMN2 mRNA, and therefore levels of full length SMN protein upon oral administration in two mouse models of SMA.

Designing Multi-target Compound Libraries with Gaussian Process Models.

We present the application of machine learning models to selecting G protein-coupled receptor (GPCR)-focused compound libraries. The library design process was realized by ant colony optimization. A proprietary Boehringer-Ingelheim reference set consisting of 3519 compounds tested in dose-response assays at 11 GPCR targets served as training data for machine learning and activity prediction. We compared the usability of the proprietary data with a public data set from ChEMBL. Gaussian process models were trained to prioritize compounds from a virtual combinatorial library. We obtained meaningful models for three of the targets (5-HT2c , MCH, A1), which were experimentally confirmed for 12 of 15 selected and synthesized or purchased compounds. Overall, the models trained on the public data predicted the observed assay results more accurately. The results of this study motivate the use of Gaussian process regression on public data for virtual screening and target-focused compound library design.

Multidimensional de†novo design reveals 5-HT2B receptor-selective ligands.

We report a multi-objective de novo design study driven by synthetic tractability and aimed at the prioritization of computer-generated 5-HT2B receptor ligands with accurately predicted target-binding affinities. Relying on quantitative bioactivity models we designed and synthesized structurally novel, selective, nanomolar, and ligand-efficient 5-HT2B modulators with sustained cell-based effects. Our results suggest that seamless amalgamation of computational activity prediction and molecular design with microfluidics-assisted synthesis enables the swift generation of small molecules with the desired polypharmacology.

Revealing the macromolecular targets of complex natural products.

Natural products have long been a source of useful biological activity for the development of new drugs. Their macromolecular targets are, however, largely unknown, which hampers rational drug design and optimization. Here we present the development and experimental validation of a computational method for the discovery of such targets. The technique does not require three-dimensional target models and may be applied to structurally complex natural products. The algorithm dissects the natural products into fragments and infers potential pharmacological targets by comparing the fragments to synthetic reference drugs with known targets. We demonstrate that this approach results in confident predictions. In a prospective validation, we show that fragments of the potent antitumour agent archazolid A, a macrolide from the myxobacterium Archangium gephyra, contain relevant information regarding its polypharmacology. Biochemical and biophysical evaluation confirmed the predictions. The results obtained corroborate the practical applicability of the computational approach to natural product 'de-orphaning'.

Combinatorial chemistry by ant colony optimization.

BACKGROUND: Prioritizing building blocks for combinatorial medicinal chemistry represents an optimization task. We present the application of an artificial ant colony algorithm to combinatorial molecular design (Molecular Ant Algorithm [MAntA]). RESULTS: In a retrospective evaluation, the ant algorithm performed favorably compared with other stochastic optimization methods. Application of MAntA to peptide design resulted in new octapeptides exhibiting substantial binding to mouse MHC-I (H-2K(b)). In a second study, MAntA generated a new functional factor Xa inhibitor by Ugi-type three-component reaction. CONCLUSION: This proof-of-concept study validates artificial ant systems as innovative computational tools for efficient building block prioritization in combinatorial chemistry. Focused activity-enriched compound collections are obtained without the need for exhaustive product enumeration.

Multi-objective molecular de novo design by adaptive fragment prioritization.

We present the development and application of a computational molecular de novo design method for obtaining bioactive compounds with desired on- and off-target binding. The approach translates the nature-inspired concept of ant colony optimization to combinatorial building block selection. By relying on publicly available structure-activity data, we developed a predictive quantitative polypharmacology model for 640 human drug targets. By taking reductive amination as an example of a privileged reaction, we obtained novel subtype-selective and multitarget-modulating dopamine D4 antagonists, as well as ligands selective for the sigma-1 receptor with accurately predicted affinities. The nanomolar potencies of the hits obtained, their high ligand efficiencies, and an overall success rate of 90 % demonstrate that this ligand-based computer-aided molecular design method may guide target-focused combinatorial chemistry.

Machine learning estimates of natural product conformational energies.

Machine learning has been used for estimation of potential energy surfaces to speed up molecular dynamics simulations of small systems. We demonstrate that this approach is feasible for significantly larger, structurally complex molecules, taking the natural product Archazolid A, a potent inhibitor of vacuolar-type ATPase, from the myxobacterium Archangium gephyra as an example. Our model estimates energies of new conformations by exploiting information from previous calculations via Gaussian process regression. Predictive variance is used to assess whether a conformation is in the interpolation region, allowing a controlled trade-off between prediction accuracy and computational speed-up. For energies of relaxed conformations at the density functional level of theory (implicit solvent, DFT/BLYP-disp3/def2-TZVP), mean absolute errors of less than 1 kcal/mol were achieved. The study demonstrates that predictive machine learning models can be developed for structurally complex, pharmaceutically relevant compounds, potentially enabling considerable speed-ups in simulations of larger molecular structures.

Combining on-chip synthesis of a focused combinatorial library with computational target prediction reveals imidazopyridine GPCR ligands.

Using the example of the Ugi three-component reaction we report a fast and efficient microfluidic-assisted entry into the imidazopyridine scaffold, where building block prioritization was coupled to a new computational method for predicting ligand-target associations. We identified an innovative GPCR-modulating combinatorial chemotype featuring ligand-efficient adenosine A1/2B and adrenergic α1A/B receptor antagonists. Our results suggest the tight integration of microfluidics-assisted synthesis with computer-based target prediction as a viable approach to rapidly generate bioactivity-focused combinatorial compound libraries with high success rates.