Search

Established molecular machine learning models process individual molecules as inputs to predict their biological, chemical, or physical properties. However, such algorithms require large datasets and have not been optimized to predict property differences between molecules. Many drug and material development tasks would benefit from an algorithm that can directly compare two molecules to guide molecular optimization and prioritization. Here, we develop DeepDelta, a pairwise deep learning approach that processes two molecules simultaneously and learns to predict property differences between two molecules from small datasets. On 10 pharmacokinetic benchmark tasks, our DeepDelta approach outperforms two established molecular machine learning algorithms, the message passing neural network (MPNN) ChemProp and Random Forest using radial fingerprints. We further analyze our performance and find that DeepDelta is particularly outperforming established approaches at predicting large differences in molecular properties and can perform scaffold hopping. Furthermore, we derive simple computational tests of our models based on mathematical invariants and show that compliance to these tests correlate with overall model performance – providing an innovative, unsupervised, and easily computable measure of expected model performance and applicability. Taken together, DeepDelta provides an accurate approach to predict molecular property differences and will allow for higher fidelity and transparency in molecular optimization for drug development and the chemical sciences.

Data subsampling is an established machine learning pre-processing technique to reduce bias in datasets. However, subsampling can lead to the removal of crucial information from the data and thereby decrease performance. Multiple different subsampling strategies have been proposed, and benchmarking is necessary to identify the best strategy for a specific machine learning task. Instead, we propose to use active machine learning as an autonomous and adaptive data subsampling strategy. We show that active learning-based subsampling can lead to better molecular machine learning performance when compared to both training models on the complete training data and 19 state-of-the-art subsampling strategies. We find that active learning is robust to errors in the data, highlighting the utility of this approach for low quality datasets. Taken together, we here describe a new, adaptive machine learning pre-processing approach and provide novel insights into the behavior and robustness of active machine learning for molecular sciences.

Monolayers of cancer-derived cell lines are widely used in the modelling of the gastrointestinal (GI) absorption of drugs and in oral drug development. However, they do not generally predict drug absorption in vivo. Here, we report a robotically handled system that uses large porcine GI tissue explants that are functionally maintained for an extended period in culture for the high-throughput interrogation (several thousand samples per day) of whole segments of the GI tract. The automated culture system provided higher predictability of drug absorption in the human GI tract than a Caco-2 Transwell system (Spearman’s correlation coefficients of 0.906 and 0.302, respectively). By using the culture system to analyse the intestinal absorption of 2,930 formulations of the peptide drug oxytocin, we discovered an absorption enhancer that resulted in a 11.3-fold increase in the oral bioavailability of oxytocin in pigs in the absence of cellular disruption of the intestinal tissue. The robotically handled whole-tissue culture system should help advance the development of oral drug formulations and might also be useful for drug screening applications. A robotically handled culture system using porcine gastrointestinal tissue explants for the high-throughput interrogation of the gastrointestinal tract predicts the absorption of oral drugs in the human gut better than Caco-2 Transwells.

Identifying potential protein-ligand interactions is central to the field of drug discovery as it facilitates the identification of potential novel drug leads, contributes to advancement from hits to leads, predicts potential off-target explanations for side effects of approved drugs or candidates, as well as de-orphans phenotypic hits. For the rapid identification of protein-ligand interactions, we here present a novel chemogenomics algorithm for the prediction of protein-ligand interactions using a new machine learning approach and novel class of descriptor. The algorithm applies Bayesian Additive Regression Trees (BART) on a newly proposed proteochemical space, termed the bow-pharmacological space. The space spans three distinctive sub-spaces that cover the protein space, the ligand space, and the interaction space. Thereby, the model extends the scope of classical target prediction or chemogenomic modelling that relies on one or two of these subspaces. Our model demonstrated excellent prediction power, reaching accuracies of up to 94.5–98.4% when evaluated on four human target datasets constituting enzymes, nuclear receptors, ion channels, and G-protein-coupled receptors . BART provided a reliable probabilistic description of the likelihood of interaction between proteins and ligands, which can be used in the prioritization of assays to be performed in both discovery and vigilance phases of small molecule development.

Drug discovery is fuelled by small-molecules, either as tools to interrogate biology or as leads for investigational therapeutics.1-3 The successful development of such entities entails a transformative power on disease modulation, but a large proportion of them fail to reach the clinic due to attrition.4 In that regard, it is increasingly recognized that a significant fraction of screening molecules present low aqueous solubility and form nano/microscale agglomerates. These colloidal aggregates can bind unspecifically to proteins, inducing local denaturation and apparent inhibition.5 While undirected interactions between small molecules and proteins are not desirable, most biological assays lack resolution to distinguish between ‘pathological’ and ‘directed’ target engagement; thereby incorrectly presenting the aggregating molecule as a positive result. Indeed, colloidal aggregates account for the largest source of false positive hits in high-throughput screens, surpassing other well-documented ‘con artists’, such as the pan-assay interference compounds

PURPOSE: A multitude of different versions of the same medication with different inactive ingredients are currently available. It has not been quantified how this has evolved historically. Furthermore, it is unknown whether healthcare professionals consider the inactive ingredient portion when prescribing medications to patients. METHODS: We used data mining to track the number of available formulations for the same medication over time and correlate the number of available versions in 2019 to the number of manufacturers, the years since first approval, and the number of prescriptions. A focused survey among healthcare professionals was conducted to query their consideration of the inactive ingredient portion of a medication during prescriptions. RESULTS: The number of available versions of a single medication have dramatically increased in the last 40 years. The number of available, different versions of medications are largely determined by the number of manufacturers producing this medication. Healthcare provides commonly do not consider these the inactive ingredient portion when prescribing a medication. CONCLUSIONS: A multitude of available versions of the same medications provides a potentially under-recognized opportunity to prescribe the most suitable formulation to a patient as a step towards personalized medicine and mitigating potential adverse events from inactive ingredients.

Active machine learning enables the automated selection of the most valuable next experiments to improve predictive modelling and hasten active retrieval in drug discovery. Although a long established theoretical concept and introduced to drug discovery approximately 15 years ago, the deployment of active learning technology in the discovery pipelines across academia and industry remains slow. With the recent re-discovered enthusiasm for artificial intelligence as well as improved flexibility of laboratory automation, active learning is expected to surge and become a key technology for molecular optimizations. This review recapitulates key findings from previous active learning studies to highlight the challenges and opportunities of applying adaptive machine learning to drug discovery. Specifically, considerations regarding implementation, infrastructural integration, and expected benefits are discussed. By focusing on these practical aspects of active learning, this review aims at providing insights for scientists planning to implement active learning workflows in their discovery pipelines.

Active machine learning is an experimental design approach that puts machine learning models in the driver seat of data acquisition and automated optimization. Introduced to drug discovery approximately 15 years ago, a handful of impressive studies have revealed the potential of active machine learning to guide molecular discovery and optimization. Most recently, researchers have shown that active learning can also be applied to datasets retrospectively, enabling automated data curation to train powerful machine learning models on small datasets. This chapter reviews the key findings from these active learning studies and summarizes remaining challenges and future opportunities of active learning as an adaptive learning technology in the context of drug discovery.

Fragment-like natural products play a pivotal role in natural product research given their improved synthetic and computational tractability as well as commercial availability compared to more complex natural product structures. A multitude of computational tools have been developed to support the generation, analysis, and application of natural fragments for drug discovery and chemical biology research. In this contribution, the challenges and opportunities in such workflows are discussed and contextualized. Multiple successful applications and validations discussed herein attest to the relevance of natural fragments for drug discovery and the utility of machine learning and data science to support such endeavors.

Nanoformulations are transforming our capacity to effectively deliver and treat a myriad of conditions. However, many nanoformulation approaches still suffer from high production complexity and low drug loading. One potential solution relies on harnessing co-assembly of drugs and small molecular excipients to facilitate nanoparticle formation through solvent exchange without the need for chemical synthesis, generating nanoparticles with up to 95% drug loading. However, there is currently no understanding which of the millions of possible combinations of small molecules can result in the formation of these nanoparticles. Here we report the development of a high-throughput screening platform coupled to machine learning to enable the rapid evaluation of such nanoformulations. Our platform identified 101 novel self-assembling drug nanoparticles from 2.1 million pairings derived from 788 candidate drugs with one of 2686 excipients, spanning treatments for multiple diseases and often harnessing well-known food additives, vitamins, or approved drugs as carrier materials – with potential for accelerated approval and translation. Given their long-term stability and potential for clinical impact, we further characterize novel sorafenib-glycyrrhizin and terbinafine-taurocholic acid nanoparticles ex vivo and in vivo. We anticipate that this platform could accelerate the development of safer and more efficacious nanoformulations with high drug loadings for a wide range of therapeutics.

Small molecule effectors are essential for drug discovery. Specific molecular recognition, reversible binding and dose-dependency are usually key requirements to ensure utility of a novel chemical entity. However, artefactual frequent-hitter and assay interference compounds may divert lead optimization and screening programmes towards attrition-prone chemical matter. Colloidal aggregates are the prime source of false positive readouts, either through protein sequestration or protein-scaffold mimicry. Nevertheless, assessment of colloidal aggregation remains somewhat overlooked and under-appreciated. In this Review, we discuss the impact of aggregation in drug discovery by analysing select examples from the literature and publicly-available datasets. We also examine and comment on technologies used to experimentally identify these potentially problematic entities. We focus on evidence-based computational filters and machine learning algorithms that may be swiftly deployed to flag chemical matter and mitigate the impact of aggregates in discovery programmes. We highlight the tools that can be used to scrutinize libraries, and identify and eliminate these problematic compounds.

The cyclodepsipeptide doliculide is a marine natural product with strong actin-polymerizing and anticancer activities. Evidence for doliculide acting as a potent and subtype-selective antagonist of prostanoid E receptor 3 (EP3) is presented. Computational target prediction suggested that this membrane receptor is a likely macromolecular target and enabled immediate in vitro validation. This proof-of-concept study demonstrates the in silico deorphanization of phenotypic screening hits as a viable concept for future natural-product-inspired chemical biology and drug discovery efforts.

Summary Optimizing reaction conditions depends on expert chemistry knowledge and laborious exploration of reaction parameters. To automate this task and augment chemical intuition, we here report a computational tool to navigate search spaces. Our approach (LabMate.ML) integrates random sampling of 0.03%–0.04% of all search space as input data with an interpretable, adaptive machine-learning algorithm. LabMate.ML can optimize many real-valued and categorical reaction parameters simultaneously, with minimal computational resources and time. In nine prospective proof-of-concept studies pursuing distinctive objectives, we demonstrate how LabMate.ML can identify optimal goal-oriented conditions for several different chemistries and substrates. Double-blind competitions and the conducted expert surveys reveal that its performance is competitive with that of human experts. LabMate.ML does not require specialized hardware, affords quantitative and interpretable reactivity insights, and autonomously formalizes chemical intuition, thereby providing an innovative framework for informed, automated experiment selection toward the democratization of synthetic chemistry.

Oral forms of medications contain "inactive" ingredients to enhance their physical properties. Using data analytics, we characterized the abundance and complexity of inactive ingredients in approved medications. A majority of medications contain ingredients that could cause adverse reactions, underscoring the need to maximize the tolerability and safety of medications and their inactive ingredients., National Instutites of Health (Grant EB000244)

Fragment-like natural products play a pivotal role in natural product research given their improved synthetic and computational tractability as well as commercial availability compared to more complex natural product structures. A multitude of computational tools have been developed to support the generation, analysis, and application of natural fragments for drug discovery and chemical biology research. In this contribution, the challenges and opportunities in such workflows are discussed and contextualized. Multiple successful applications and validations discussed herein attest to the relevance of natural fragments for drug discovery and the utility of machine learning and data science to support such endeavors.

High-throughput and high-content screening campaigns have resulted in the creation of large chemogenomic matrices. These matrices form the training data which is used to build ligand-target interaction models for pharmacological and chemical biology research. While academic, government, and industrial efforts continuously add to the ligand-target data pairs available for modeling, major research efforts are devoted to improving machine learning techniques to cope with the sparseness, heterogeneity, and size of available datasets as well as inherent noise and bias. This "race of arms" has led to the creation of algorithms to generate highly complex models with high prediction performance at the cost of training efficiency as well as interpretability.In contrast, recent studies have challenged the necessity for "big data" in chemogenomic modeling and found that models built on larger numbers of examples do not necessarily result in better predictive abilities. Automated adaptive selection of the training data (ligand-target instances) used for model creation can result in considerably smaller training sets that retain prediction performance on par with training using hundreds of thousands of data points. In this chapter, we describe the protocols used for one such iterative chemogenomic selection technique, including model construction and update as well as possible techniques for evaluations of constructed models and analysis of the iterative model construction.

Monolayers of cancer-derived cell lines are widely used in the modelling of the gastrointestinal (GI) absorption of drugs and in oral drug development. However, they do not generally predict drug absorption in vivo. Here, we report a robotically handled system that uses large porcine GI tissue explants that are functionally maintained for an extended period in culture for the high-throughput interrogation (several thousand samples per day) of whole segments of the GI tract. The automated culture system provided higher predictability of drug absorption in the human GI tract than a Caco-2 Transwell system (Spearman's correlation coefficients of 0.906 and 0.302, respectively). By using the culture system to analyse the intestinal absorption of 2,930 formulations of the peptide drug oxytocin, we discovered an absorption enhancer that resulted in a 11.3-fold increase in the oral bioavailability of oxytocin in pigs in the absence of cellular disruption of the intestinal tissue. The robotically handled whole-tissue culture system should help advance the development of oral drug formulations and might also be useful for drug screening applications.

Active machine learning puts artificial intelligence in charge of a sequential, feedback-driven discovery process., Active machine learning puts artificial intelligence in charge of a sequential, feedback-driven discovery process. We present the application of a multi-objective active learning scheme for identifying small molecules that inhibit the protein–protein interaction between the anti-cancer target CXC chemokine receptor 4 (CXCR4) and its endogenous ligand CXCL-12 (SDF-1). Experimental design by active learning was used to retrieve informative active compounds that continuously improved the adaptive structure–activity model. The balanced character of the compound selection function rapidly delivered new molecular structures with the desired inhibitory activity and at the same time allowed us to focus on informative compounds for model adjustment. The results of our study validate active learning for prospective ligand finding by adaptive, focused screening of large compound repositories and virtual compound libraries.

Automatisiertes De-novo-Molekuldesign fuhrte zur Entdeckung eines neuartigen Inhibitors der Death-Associated Proteinkinase 3 (DAPK3). Eine erste Kristallstruktur der inaktiven DAPK3 als Homodimer im Komplex mit dem fragmentartigen Liganden zeigt dessen Bindung in der ATP-Bindetasche. Das verwendete maschinelle Lernverfahren hat neben DAPK3 weitere Targets korrekt vorhergesagt, sowohl fur den entworfenen Liganden als auch fur den strukturell verwandten und bereits auf dem Markt befindlichen Wirkstoff Azosemid. Die vorliegende Studie bestatigt das Konzept des computergestutzten ligandenbasierten De-novo-Molekuldesigns als geeignet fur die Erzeugung neuartiger Liganden und potentieller Startpunkte fur die Wirkstofffindung.

Fragment-like natural products were identified as ligand-efficient chemical matter for hit-to-lead development and chemical-probe discovery. Relying on a computational method using a topological pharmacophore descriptor and a drug database, several macromolecular targets from distinct protein families were expeditiously retrieved for structurally unrelated chemotypes. The selected fragments feature structural dissimilarity to the reference compounds and suitable target affinity, and they offer opportunities for chemical optimization. Experimental confirmation of hitherto unknown macromolecular targets for the selected molecules corroborate the usefulness of the computational approach and suggests broad applicability to chemical biology and molecular medicine.

High-throughput and high-content screening campaigns have resulted in the creation of large chemogenomic matrices. These matrices form the training data which is used to build ligand-target interaction models for pharmacological and chemical biology research. While academic, government, and industrial efforts continuously add to the ligand-target data pairs available for modeling, major research efforts are devoted to improving machine learning techniques to cope with the sparseness, heterogeneity, and size of available datasets as well as inherent noise and bias. This "race of arms" has led to the creation of algorithms to generate highly complex models with high prediction performance at the cost of training efficiency as well as interpretability.In contrast, recent studies have challenged the necessity for "big data" in chemogenomic modeling and found that models built on larger numbers of examples do not necessarily result in better predictive abilities. Automated adaptive selection of the training data (ligand-target instances) used for model creation can result in considerably smaller training sets that retain prediction performance on par with training using hundreds of thousands of data points. In this chapter, we describe the protocols used for one such iterative chemogenomic selection technique, including model construction and update as well as possible techniques for evaluations of constructed models and analysis of the iterative model construction.

We present the application of the generative topographic map algorithm to visualize the chemical space populated by natural products and synthetic drugs. Generative topographic maps may be used for nonlinear dimensionality reduction and probabilistic modeling. For compound mapping, we represented the molecules by two-dimensional pharmacophore features (chemically advanced template search descriptor). The results obtained suggest a close resemblance of synthetic drugs with natural products in terms of their pharmacophore features, despite pronounced differences in chemical structure. Generative topographic map-based cluster analysis revealed both known and new potential activities of natural products and drug-like compounds. We conclude that the generative topographic map method is suitable for inferring functional similarities between these two classes of compounds and predicting macromolecular targets of natural products.

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.

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'.

Epigenetic effects are exerted by a variety of factors and evidence increases that common drugs such as opioids, cannabinoids, valproic acid, or cytostatics may induce alterations in DNA methylation patterns or histone conformations. These effects occur via chemical structural interactions with epigenetic enzymes, through interactions with DNA repair mechanisms. Computational predictions indicate that one-twentieth of all drugs might potentially interact with human histone deacetylase, which was prospectively experimentally verified for the compound with the highest predicted interaction probability. These epigenetic effects add to wanted and unwanted drug effects, contributing to mechanisms of drug resistance or disease-related and unrelated phenotypes. Because epigenetic changes might be transmitted to offspring, the need for reliable and cost-effective epigenetic screening tools becomes acute.

Aim: Computational chemogenomics models the compound–protein interaction space, typically for drug discovery, where existing methods predominantly either incorporate increasing numbers of bioactivity samples or focus on specific subfamilies of proteins and ligands. As an alternative to modeling entire large datasets at once, active learning adaptively incorporates a minimum of informative examples for modeling, yielding compact but high quality models. Results/methodology: We assessed active learning for protein/target family-wide chemogenomic modeling by replicate experiment. Results demonstrate that small yet highly predictive models can be extracted from only 10–25% of large bioactivity datasets, irrespective of molecule descriptors used. Conclusion: Chemogenomic active learning identifies small subsets of ligand–target interactions in a large screening database that lead to knowledge discovery and highly predictive models.

Cancer stem cells (CSCs) play major roles in cancer initiation, metastasis, recurrence and therapeutic resistance. Targeting CSCs represents a promising strategy for cancer treatment. The purpose of this study was to identify selective inhibitors of breast CSCs (BCSCs). We carried out a cell-based phenotypic screening with cell viability as a primary endpoint, using a collection of 2,546 FDA-approved drugs and drug-like molecules in spheres formed by malignant human breast gland-derived cells (HMLER-shEcad cells, representing BCSCs) and control immortalized non-tumorigenic human mammary cells (HMLE cells, representing normal stem cells). 19 compounds were identified from screening. The chemically related molecules benztropine mesylate and deptropine citrate were selected for further validation and both potently inhibited sphere formation and self-renewal of BCSCs in vitro. Benztropine mesylate treatment decreased cell subpopulations with high ALDH activity and with a CD44+/CD24- phenotype. In vivo, benztropine mesylate inhibited tumor-initiating potential in a 4T1 mouse model. Functional studies indicated that benztropine mesylate inhibits functions of CSCs via the acetylcholine receptors, dopamine transporters/receptors, and/or histamine receptors. In summary, our findings identify benztropine mesylate as an inhibitor of BCSCs in vitro and in vivo. This study also provides a screening platform for identification of additional anti-CSC agents., Oncotarget, 8 (1), ISSN:1949-2553

Drug discovery is driven by the identification of new chemical entities (NCEs).1,2 Virtual screening and de novo design techniques have been proven to serve this purpose, thereby complementing experimental biochemical and biological approaches.3 Still, it remains a matter of debate, which particular molecular representation and similarity index are preferable for a given drug target in order to identify appropriate NCEs with minimal synthetic and testing effort involved.4 Ligand-based chemical similarity approaches have been effectively applied to large-scale activity and target prediction for known drugs, some of the prominent methods being PASS developed by Poroikov et al.,5 the techniques conceived by Mestres and co-workers,6 and the similarity ensemble approach (SEA) implemented by the Shoichet group.7 Here, we compared several popular two-dimensional molecular representations for their ability to retrieve actives (enrichment potential) and chemotypes (scaffold-hopping potential) from a collection of druglike bioactive compounds. Subsequently the applied chemical advanced template search (CATS)8 was applied to predicting potential drug targets for a virtually assembled combinatorial compound library, from which we synthesized and successfully tested candidate compounds. The results demonstrate that CATS is not only suited for its intended purpose of NCE retrieval by scaffold-hopping,9 but also for reliable target profiling of ‘orphan’ virtual molecules.10 It thereby complements the suite of available validated tools for target prediction.

Natural products and their molecular frameworks have a long tradition as valuable starting points for medicinal chemistry and drug discovery. Recently, there has been a revitalization of interest in the inclusion of these chemotypes in compound collections for screening and achieving selective target modulation. Here we discuss natural-product-inspired drug discovery with a focus on recent advances in the design of synthetically tractable small molecules that mimic nature's chemistry. We highlight the potential of innovative computational tools in processing structurally complex natural products to predict their macromolecular targets and attempt to forecast the role that natural-product-derived fragments and fragment-like natural products will play in next-generation drug discovery.