Machine Learning for Discovery of New ADORA Modulators

A non-purine inhibitor of xanthine oxidoreductase mitigates adenosine triphosphate degradation under hypoxic conditions in mouse brain

The brain is an organ that consumes a substantial amount of oxygen, and a reduction in oxygen concentration can rapidly lead to significant and irreversible brain injury. The progression of brain injury during hypoxia involves the depletion of intracellular adenosine triphosphate (ATP) due to decreased oxidative phosphorylation in the inner mitochondrial membrane. Allopurinol is a purine analog inhibitor of xanthine oxidoreductase that protects against hypoxic/ischemic brain injury; however, its underlying mechanism of action remains unclear. In addition, febuxostat is a non-purine xanthine oxidoreductase inhibitor with a different inhibitory mechanism from allopurinol. The impact of febuxostat on brain injury has not been well investigated. Therefore, this study aimed to examine brain ATP and its catabolite levels in the presence or absence of allopurinol and febuxostat under hypoxic conditions by inactivating brain metabolism using focal microwave irradiation. The hypoxic treatment caused a decrease in the adenylate energy charge and ATP levels and an increase in its catabolic products in mouse brains. The febuxostat group showed higher energy charge and ATP levels and lower ATP catabolites than the control group. Notably, despite the comparable suppression of uric acid production in both inhibitor groups, allopurinol treatment was less effective than febuxostat. These results suggest that febuxostat effectively prevents hypoxia-induced ATP degradation in the brain and that its effect is more potent than allopurinol. This study will contribute to developing therapies for improving hypoxia-induced brain dysfunction.

Repurposing lapatinib as a triple antagonist of chemokine receptors 3, 4, and 5

Chemokine receptors CCR3, CCR4, and CCR5 are G protein-coupled receptors implicated in diseases like cancer, Alzheimer's, asthma, human immunodeficiency virus (HIV), and macular degeneration. Recently, CCR3 and CCR4 have emerged as potential stroke targets. Although only the CCR5 antagonist maraviroc is US Food and Drug Administration-approved (for HIV), we curated data on CCR3, CCR4, and CCR5 antagonists from ChEMBL to develop and validate machine learning models. The top 5-fold cross-validation statistics for these models were high for both classification and regression models for CCR3 (receiver operating characteristic [ROC], 0.94; R2 = 0.8), CCR4 (ROC, 0.98; R2 = 0.57), and CCR5 (ROC, 0.96; R2 = 0.78). The models for CCR3/4 were used to screen a small library of US Food and Drug Administration-approved drugs and 17 were initially tested in vitro against both CCR3/4 receptors. A promising compound lapatinib, a dual tyrosine kinase inhibitor, was identified as an antagonist for CCR3 (IC50, 0.7 μM) and CCR4 (IC50, 1.8 μM). Additional testing also identified it as an CCR5 antagonist (IC50, 0.9 μM), and it showed moderate in vitro HIV I inhibition. We demonstrated how machine learning can be used to identify molecules for repurposing as antagonists for G protein-coupled receptors such as CCR3, CCR4, and CCR5. Lapatinib may represent a new orally available chemical probe for these 3 receptors, and it provides a starting point for further chemical optimization for multiple diseases impacting human health. SIGNIFICANCE STATEMENT: We describe the building of machine learning models for the chemokine receptors CCR3, CCR4, and CCR5 trained on data from the ChEMBL database. Using these models, we identified lapatinib as a potent inhibitor of CCR3, CCR4, and CCR5. Our study illustrates the potential of machine learning in identifying molecules for repurposing as antagonists for G protein-coupled receptors, including CCR3, CCR4, and CCR5, which have various therapeutic applications.

Machine learning-aided search for ligands of P2Y6 and other P2Y receptors

The P2Y6 receptor, activated by uridine diphosphate (UDP), is a target for antagonists in inflammatory, neurodegenerative, and metabolic disorders, yet few potent and selective antagonists are known to date. This prompted us to use machine learning as a novel approach to aid ligand discovery, with pharmacological evaluation at three P2YR subtypes: initially P2Y6 and subsequently P2Y1 and P2Y14. Relying on extensive published data for P2Y6R agonists, we generated and validated an array of classification machine learning model using the algorithms deep learning (DL), adaboost classifier (ada), Bernoulli NB (bnb), k-nearest neighbors (kNN) classifier, logistic regression (lreg), random forest classifier (rf), support vector classification (SVC), and XGBoost (XGB) classifier models, and the common consensus was applied to molecular selection of 21 diverse structures. Compounds were screened using human P2Y6R-induced functional calcium transients in transfected 1321N1 astrocytoma cells and fluorescent binding inhibition at closely related hP2Y14R expressed in CHO cells. The hit compound ABBV-744, an experimental anticancer drug with a 6-methyl-7-oxo-6,7-dihydro-1H-pyrrolo[2,3-c]pyridine scaffold, had multifaceted interactions with the P2YR family: hP2Y6R inhibition in a non-surmountable fashion, suggesting that noncompetitive antagonism, and hP2Y1R enhancement, but not hP2Y14R binding inhibition. Other machine learning-selected compounds were either weak (experimental anti-asthmatic drug AZD5423 with a phenyl-1H-indazole scaffold) or inactive in inhibiting the hP2Y6R. Experimental drugs TAK-593 and GSK1070916 (100 µM) inhibited P2Y14R fluorescent binding by 50% and 38%, respectively, and all other compounds by < 20%. Thus, machine learning has led the way toward revealing previously unknown modulators of several P2YR subtypes that have varied effects.

Introduction of a single-nucleotide variant, rs16851030, into the ADORA1 gene increased cellular susceptibility to hypoxia

Aim: Rs16851030, a single-nucleotide variant located in the 3'-untranslated region of the ADORA1 gene, has been proposed as a potential marker of caffeine sensitivity in apnea of prematurity. Besides, it is associated with aspirin-induced asthma and the development of acute chest syndrome. However, its functional significance is still unconfirmed. This study aimed to elucidate the functional impact of rs16851030 by using CRISPR/Cas9 approach to induce the DNA variant and attendant physiological changes.Methods: Rs16851030 was introduced into HEK293 cells via homology-directed repair (HDR). Edited cells were fluorescence-enriched, sorted, isolated, and expanded into single-cell-derived clones. The edit was confirmed by Sanger sequencing. RNA sequencing was used to analyze affected pathways.Results: Rs16851030-mutant cells showed increased susceptibility to hypoxia, a condition related to apnea of prematurity. After 24 h of hypoxia, the viability of mutant clones 1 and 2 was low compared with wild-type cells (75.45% and 74.47% vs. 96.34%). RNA sequencing revealed transcriptomic changes linked to this increased vulnerability.Conclusion: Rs16851030 impairs cellular resistance to hypoxia, suggesting its role in conditions like apnea of prematurity. Further research should investigate the molecular mechanisms and transcriptomic alterations caused by rs16851030 under hypoxic conditions.

The Goldilocks paradigm: comparing classical machine learning, large language models, and few-shot learning for drug discovery applications

Recent advances in machine learning (ML) have led to newer model architectures including transformers (large language models, LLMs) showing state of the art results in text generation and image analysis as well as few-shot learning (FSLC) models which offer predictive power with extremely small datasets. These new architectures may offer promise, yet the 'no-free lunch' theorem suggests that no single model algorithm can outperform at all possible tasks. Here, we explore the capabilities of classical (SVR), FSLC, and transformer models (MolBART) over a range of dataset tasks and show a 'goldilocks zone' for each model type, in which dataset size and feature distribution (i.e. dataset "diversity") determines the optimal algorithm strategy. When datasets are small ( < 50 molecules), FSLC tend to outperform both classical ML and transformers. When datasets are small-to-medium sized (50-240 molecules) and diverse, transformers outperform both classical models and few-shot learning. Finally, when datasets are of larger and of sufficient size, classical models then perform the best, suggesting that the optimal model to choose likely depends on the dataset available, its size and diversity. These findings may help to answer the perennial question of which ML algorithm is to be used when faced with a new dataset.

New paradigms in purinergic receptor ligand discovery

The discovery and clinical implementation of modulators of adenosine, P2Y and P2X receptors (comprising nineteen subtypes) have progressed dramatically in ∼50 years since Burnstock's definition of purinergic signaling. Although most clinical trials of selective ligands (agonists and antagonists) of certain purinergic receptors failed, there is a renewed impetus to redirect efforts to new disease conditions and the discovery of more selective or targeted compounds with potentially reduced side effects, such as biased GPCR agonists. The elucidation of new receptor and enzyme structures is steering rational design of potent and selective agonists, antagonists, allosteric modulators and inhibitors. A2A adenosine receptor (AR) antagonists are being applied to neurodegenerative conditions and cancer immunotherapy. A3AR agonists have potential for treating chronic inflammation (e.g. psoriasis), stroke and pain, as well as cancer. P2YR modulators are being considered for treating inflammation, metabolic disorders, acute kidney injury, cancer, pain and other conditions, often with an immune mechanism. ADP-activated P2Y12R antagonists are widely used as antithrombotic drugs, while their repurposing toward neuroinflammation is considered. P2X3 antagonists have been in clinical trials for chronic cough. P2X7 antagonists have been in clinical trials for inflammatory diseases and depression (compounds that penetrate the blood-brain barrier). Thus, purinergic signaling is now recognized as an immense regulatory system in the body for rebalancing tissues and organs under stress, which can be adjusted by drug intervention for therapeutic purposes. The lack of success of many previous clinical trials can be overcome given more advanced pharmacokinetic and pharmacodynamic approaches, including structure-based drug design, prodrugs and biased signaling. This article is part of the Special Issue on "Purinergic Signaling: 50 years".

To Affinity and Beyond: A Personal Reflection on the Design and Discovery of Drugs

Faced with new and as yet unmet medical need, the stark underperformance of the pharmaceutical discovery process is well described if not perfectly understood. Driven primarily by profit rather than societal need, the search for new pharmaceutical products—small molecule drugs, biologicals, and vaccines—is neither properly funded nor sufficiently systematic. Many innovative approaches remain significantly underused and severely underappreciated, while dominant methodologies are replete with problems and limitations. Design is a component of drug discovery that is much discussed but seldom realised. In and of itself, technical innovation alone is unlikely to fulfil all the possibilities of drug discovery if the necessary underlying infrastructure remains unaltered. A fundamental revision in attitudes, with greater reliance on design powered by computational approaches, as well as a move away from the commercial imperative, is thus essential to capitalise fully on the potential of pharmaceutical intervention in healthcare.