Drug discovery remains constrained by high attrition rates and the fragmented evaluation of exposure, efficacy, and safety. Mechanistic models offer a biologically grounded framework for connecting these determinants across multiple levels of biological organization. This may help improve translational decision-making by supporting earlier and more integrated assessment of candidate progression. This narrative review examines the conceptual basis and current role of next-generation mechanistic models in drug discovery, with emphasis on physiologically based pharmacokinetic models, virtual cell-based assays, quantitative systems pharmacology, artificial intelligence (AI)-augmented mechanistic models, and emerging virtual-cell frameworks. It highlights how these approaches may connect efficacy and safety across biological scales, support in vitro-to-in vivo extrapolation, incorporate in silico predictions, and improve candidate prioritization. The literature was surveyed through PubMed searches conducted up to 25 May 2026. Next-generation mechanistic models are unlikely to transform drug discovery simply by increasing biological detail or computational sophistication. Progress in this direction will depend on standardized data streams, robust validation, explicit model calibration, reproducibility, tighter integration between models, and careful alignment between model design and context of use. Under these conditions, mechanistic frameworks may become important components of a more predictive and less attrition-prone drug discovery pipeline.
Mitochondrial safety assessment is used in drug discovery, supported by bioenergetic profiling, mechanistic assays, human-relevant cellular systems, and multidimensional data. These advances have improved detection of mitochondrial perturbation but have not solved the harder problem: how such signals should be interpreted and translated into discovery decisions. This perspective proposes a qualitative decision-centered framework for interpreting mitochondrial findings. This framework is anchored in reserve-demand biology, which explains why mitochondrial perturbations become consequential when drug-induced reductions in bioenergetic capacity intersect with tissue-specific demand, exposure, duration, and stress context. Furthermore, the authors describe a qualitative Translational Risk Profile organized around mechanistic severity, exposure relevance, temporal progression, and translational concordance. This profile is paired with a Decision Taxonomy: Stop, Optimize, Monitor, or Acceptable Risk. Examples illustrate how mitochondrial evidence patterns can support different discovery actions. The major limitation in mitochondrial safety assessment is no longer signal detection, but decision-oriented interpretation. Future progress will depend on integrating mechanism, exposure, duration, biomarkers, human-relevant models and quantitative or computational evidence into explicit decision frameworks. Mitochondrial findings should not be treated as binary hazards. They should be interpreted as context-dependent evidence that can guide chemistry, candidate selection, monitoring strategy, and translational risk management.
Molecular Skeleton Editing has emerged as a powerful strategy in medicinal chemistry, gaining significant attention over the past decade for its ability to rapidly access structurally diverse molecules. This approach encompasses the substitution, deletion and insertion of atoms within molecular frameworks, often resulting in ring expansion or contraction. From a drug discovery perspective, these transformations enable the direct conversion of core scaffolds into bioisosteric analogues while preserving substitution patterns, providing an efficient route to optimize biological and physicochemical properties. This review focuses on molecular editing approaches relevant to drug discovery that involve single-atom substitution, deletion or insertion, leading to skeletal modifications of cyclic frameworks. Representative examples highlighting the synthesis of therapeutically relevant compounds through these methodologies will be discussed. Peripheral editing, skeletal recasting, stereochemical editing, and functional group transpositions are beyond the scope of this review. Recent advances demonstrate the considerable potential of molecular skeletal editing to accelerate lead optimization by enabling direct scaffold diversification from existing drugs and natural products, avoiding de novo synthesis. Despite these advances, important challenges remain, particularly the extension of editing methodologies to saturated sp³-rich systems and the development of stereoselective processes that preserve or generate optically pure products.
Despite remarkable advances in computational methods, pre-clinical drug discovery continues to grapple with rising timelines and costs. Software, data, and automation are more powerful than ever, and increasingly these technologies are being embraced. However, there is still work to be done to translate this potential into meaningful reductions in cost and time. This perspective discusses the growth in drug discovery capability, exploring modern data infrastructure including cloud-native platforms, active learning, and laboratory automation. It covers emerging technologies such as LLM-based orchestration and emulation. Implementation examples illustrate successes and challenges. AI presents an opportunity to envisage a new approach to drug discovery, but cultural and technological changes are required. The exponential growth in computational drug discovery tools requires solutions that enable researchers to access scalable and robust capabilities more easily. Data generation is usually the slowest and most expensive part of the design cycle; we advocate for a rigorous application of statistical methods focussing on learning efficiency from data over absolute predictive accuracy of models. Automation also plays a critical role in enabling rapid, high-quality data generation. Focussing on modular interoperable automated units with more attractive economics will drive much wider adoption.
High failure rates in clinical drug development are often attributed to inadequate therapeutic efficacy. While binding affinity has traditionally guided lead optimization, it reflects only the equilibrium state of drug-target interactions and often correlates poorly with in vivo pharmacological responses. This limitation has prompted growing interest in kinetic parameters that more accurately capture the dynamic nature of drug-target interactions. This review focuses on drug-target residence time (τ), defined as the reciprocal of the ligand dissociation rate constant (koff), which has emerged as a crucial determinant of drug efficacy. The authors discuss the impact of residence time on pharmacological outcomes, summarize factors influencing residence time, and outline experimental and computational approaches for its evaluation. This review is based on literature searches conducted using PubMed and Web of Science to identify articles published between the 2000 to 2025. Integrating residence time and traditional binding affinity provides a more comprehensive framework for understanding drug-target interactions and guiding rational drug design. Optimizing residence time can enhance pharmacodynamic efficacy, improve target selectivity, and enhance safety. Accordingly, residence time is emerging as a key kinetic parameter in modern drug discovery.
GitHub has become essential to AI-driven drug discovery by facilitating code sharing, collaboration, and reproducible workflows. As more tools for screening, modeling, and data-driven decision-making are hosted on GitHub, researchers need clear, rigorous methods to identify, evaluate, and reuse repositories in a scientifically robust manner. In this report, the authors summarize GitHub concepts most relevant to research practice (e.g. repositories, documentation, licensing, releases, testing, and archiving) and propose a practical framework for navigating drug-discovery repositories. The report incorporates a Scopus trend analysis from 2013 to 2024 using a GitHub-focused search with drug discovery keywords, as well as keyword-based counts of GitHub repositories in selected categories to show topic density. GitHub repositories should be recognized as peer-assessable research outputs in AI-driven drug discovery rather than as supplementary material. At a minimum, this means clear documentation of intended use and limitations, a pinned environment or container for reproducibility, an explicit license, and automated testing via continuous integration. In addition, enhanced validation and governance are necessary to bridge exploratory research code with translational and industrial reliability expectations.
Amyotrophic lateral sclerosis (ALS) is a fatal neurodegenerative disease characterized by progressive motor neuron loss, with limited therapeutic options. Among the few approved drugs, edaravone, a free radical scavenger developed originally for ischemic stroke, has attracted particular attention for its ability to counteract oxidative stress, a key driver of neurodegeneration. Its amphipathic structure and ability to cross the blood-brain barrier support its potential neuroprotective action. The authors discuss preclinical studies demonstrating edaravone's ability to reduce oxidative damage, preserve mitochondrial function, and modulate neuroinflammatory responses in ALS cellular and animal models. They discuss variations in dosage, timing, and disease models that produced heterogeneous results. In transgenic mice, edaravone may delay symptom onset and modestly extend survival, but these effects are inconsistent and often limited to early disease stages. Clinically, edaravone provides modest benefits in a subset of patients, reflecting the translational gap between preclinical efficacy and clinical relevance. This case highlights broader challenges in ALS drug discovery, including limited model predictivity, methodological variability, and lack of patient stratification. The edaravone experience highlights key lessons for future neuroprotective approaches: the importance of standardized preclinical design, integration of human-based models, early pharmacokinetic validation, and biomarker-driven trials to advance precision neuroprotection in ALS.
The slow pipeline of antimicrobial drug development stands in stark contrast to the continued expansion of microbial infections, which pose a persistent and major threat to global public health. Traditional discovery strategies, including natural product extraction, structural modification and high-throughput screening, are limited by low efficiency, high costs and slow innovation. Artificial intelligence (AI), particularly machine learning, deep learning, and natural language processing, is now reshaping drug discovery, bringing unprecedented speed and precision to the development of novel antimicrobial drugs. This review summarizes recent advances and challenges in AI-driven strategies for discovering antimicrobial drugs against bacterial, fungal, and viral infections, covering major drug classes including small molecules, peptides, phages, and protein drugs. The article was based on literature retrieved through a structured, iterative search of major scientific databases (PubMed, Web of Science, Scopus), with a focus on studies published between 2020 and 2025. The integration of AI-driven predictive and generative strategies will be the defining cornerstone of the next decade's 'autonomous discovery' paradigm for antimicrobial drug development, despite current enduring challenges including data bias, lack of standardized benchmarking frameworks, and clinical translational gaps.
Prion diseases are rare, fatal neurodegenerative disorders caused by the conformational conversion of cellular prion protein (PrP) into a pathogenic counterpart called PrPSc, that replicates by templated misfolding. The resulting misfolding cascade leads to synaptic failure and neuronal loss. Despite major advances in understanding prion structure and biology, no therapies exist. The mechanism of prion propagation challenges conventional drug discovery, limiting targets. Small molecules, immunotherapies, and gene silencing, have shown limited clinical success. Increasingly, research is shifting toward alternative strategies to modulate proteostasis, enhance clearance, or target transient folding and misfolding intermediates of PrP, offering new therapeutic opportunities. However, these emerging strategies remain at an early conceptual or preclinical stage, and significant translational hurdles remain. This chapter reviews prion drug discovery through the lens of PrP energy landscape. It summarizes efforts to target PrP conversion and highlights emerging approaches focused on degradation and intermediate targeting. Advances in structural biology and computational modelling are discussed as tools to identify novel therapeutic vulnerabilities. The persistent failure of conventional strategies underscores the need for innovative approaches. Targeting transient conformational intermediates and exploiting cellular quality-control systems may redefine druggability in prion diseases and open new avenues for effective intervention.
AI has tremendous potential to reduce time and costs taken to discover and develop new medical entities. As technology evolves, it is essential to learn from successes and failures to realign expectations for scientists, stakeholders and investors. The authors discuss the challenges associated with the traditional reductionist approach to drug discovery which relies on incomplete data for target validation and, specifically for small molecules, the expanse of chemical space providing potential candidates. The promise of AI is illustrated by both early success and failure stories. Lessons learned are provided at levels of realism, adoption and integration of AI within current Research and Development (R&D) organizational structures. The first decade of AI adoption in Big BioPharma has been characterized by genuine breakthroughs and sobering realities. While AI has delivered notable accelerations in hit identification and early-stage design, it has yet to fundamentally alter the success rates of late-stage clinical trials. The industry has learned that AI is neither a silver bullet nor a passing fad, though a critical and evolving component of modern R&D. By consolidating lessons from early adoption, the next decade may see AI truly shift the innovation frontier in global pharmaceutical discovery.
Vision loss in older adults is largely driven by age-related macular degeneration (AMD), characterized by progressive central visual field damage and functional decline. While current options for wet and dry AMD are limited and expensive, drug repurposing represents a promising strategy to accelerate the discovery of effective, accessible treatment by leveraging medications with established safety profiles. Notably, anti-diabetic agents including metformin, sulfonylureas, glucagon-like peptide-1 receptor agonists (GLP-1RAs), and insulin have emerged as modulators of the retinal pigment epithelium (RPE) function, photoreceptors, and retinal vascular integrity. This review highlights the roles of oxidative stress, inflammation, and complement-mediated immune dysregulation in AMD pathogenesis, alongside preclinical data demonstrating metformin's protective effects via AMP-activated protein kinase (AMPK) activation. Population-based studies and meta-analyses further suggest a modest reduction in AMD risk associated with metformin use in both diabetic and non-diabetic cohorts. Additional pharmacological agents include statins, glyburide, L-DOPA, fluoxetine, dimethyl fumarate, and nutraceuticals such as curcumin, melatonin, and N-acetylcysteine. Early AMD prevention through repurposed therapeutics, guided by AI-driven design and systems biology, may enable personalized care via multimodal risk stratification incorporating genetic, metabolomic, and microbiome data. Rigorous, stratified clinical trials integrating bioinformatics and precision medicine are essential to validate the most effective candidates.
β-Carbonic anhydrases (β-CAs) are zinc-dependent metalloenzymes that catalyze the reversible hydration of carbon dioxide to bicarbonate and protons. They are widely distributed in bacteria, where they support pH regulation, inorganic carbon homeostasis, and central metabolism. Unlike humans, which express only α-class carbonic anhydrases, many bacterial pathogens encode β-CAs, highlighting these enzymes as attractive antibacterial targets with reduced risk of host cross-reactivity. This review discusses integrated in silico and in vitro strategies for the discovery and validation of small-molecule inhibitors targeting bacterial β-CAs. Computational approaches - including pharmacophore modeling, molecular docking, molecular dynamics simulations, and machine learning - are increasingly used to prioritize and optimize candidate inhibitors. Experimental validation employs enzymatic activity assays, biophysical binding techniques, and whole-cell assays to assess target engagement and antibacterial effects. Current inhibitor classes include sulfonamides, coumarins, dithiocarbamates, phenolic compounds, and natural products, with selected chemotypes demonstrating antibacterial or antivirulence activity in specific models. Relevant literature was identified through searches of PubMed, Web of Science, and Scopus, focusing on studies published between approximately 2000 and 2025. β-CAs represent a tractable yet underexploited antibacterial target class. Successful translation will depend on improving bacterial penetration, pharmacokinetics, and target engagement. When strategically positioned as adjunctive or context-dependent therapies, β-CA inhibitors may contribute to the treatment of drug-resistant bacterial infections, including tuberculosis.
ALS drug discovery has long depended on model systems that incompletely capture human disease heterogeneity, aging, and TDP-43 proteinopathy. Patient-derived platforms have therefore emerged as increasingly important human-relevant complements to animal and molecular models. This Critical Perspective examines when patient-derived ALS models genuinely change therapeutic decision-making rather than merely add mechanistic insight. The authors then propose a heuristic framework based on disease-relevant phenotype recapitulation, capture of patient-to-patient heterogeneity, and generation of findings that influence therapeutic prioritization or clinical translation. Furthermore, the authors evaluate iPSC-derived motor neurons, directly reprogrammed neurons, glial co-cultures, organoids, neural networks, and organ-chip systems against these conditions, while also addressing aging fidelity, reproducibility, upper motor neuron modeling, and regulatory implementation. Patient-derived models are not yet standalone decision-grade tools for ALS drug development. Their present value lies in functioning as a human-biology filter for target discovery, reverse translation, biomarker development, and patient stratification when used within rigorous, standardized, and clinically linked workflows. The strongest current evidence supports proof-of-principle rather than generalized predictive validity.
Methicillin-resistant Staphylococcus aureus (MRSA) is a leading cause of severe infections with excess mortality. Progress with traditional antibiotics has been incremental, while resistance, persistence, tolerance, and biofilm formation continue to erode effectiveness. Parallel advances in small-molecule discovery, long-acting lipoglycopeptides, next-generation β-lactams, and non-traditional modalities such as bacteriophage lysins have renewed interest in expanding therapeutic options, though transition from promising preclinical signals to clinical benefit remains challenging. A literature search was conducted using PubMed/MEDLINE, and Embase, for articles published from January 2010 through March 2025. This review synthesizes developments across: (i) agents in key clinical trials for invasive MRSA infection, emphasizing on trial designs, efficacy, and safety considerations; (ii) clinical study data with newer agents for MRSA skin infections and their potential application in invasive disease; (iii) preclinical pipelines including natural products, novel compounds, and other innovative antimicrobial strategies. Among investigated agents, ceftobiprole, ceftaroline, dalbavancin, and exebacase represent promising options for invasive MRSA infections. The pipeline is further strengthened by novel classes and antimicrobial peptides, which show anti-MRSA activity and a low risk for resistance in preclinical models. Continued multidisciplinary collaboration and robust clinical trial infrastructure are essential to translate these advances into improved patient outcomes.
Structural biology has become a cornerstone of modern drug discovery, enabling atomic-level insights into protein - ligand interactions and guiding rational therapeutic design. As the field evolves, it faces growing demands for accuracy, reproducibility, and integration with computational and pharmacological data. This article explores the impact of sample heterogeneity and radiation damage on macromolecular crystallography, emphasizing how these factors can compromise structural integrity. It reviews current strategies for mitigating crystal damage, including optimized cooling, dose-aware data collection, and emerging technologies such as serial crystallography and advanced detectors. The manuscript also discusses the limitations of existing validation tools and the need for improved metadata reporting to ensure reliable structural models. Cryo-electron tomography is highlighted as a promising technique for studying drug - target interactions in native cellular environments, offering complementary insights to traditional crystallographic methods. To advance drug discovery, the structural biology community must adopt unified standards for data validation and experimental documentation. High-quality, reproducible structures are essential for minimizing artifacts and supporting AI-driven modeling and screening. A coordinated effort to integrate damage-aware practices and metadata standards will enhance the fidelity of structural data and its utility in therapeutic innovation.
Affinity selection mass spectrometry (AS-MS) is a powerful label-free technique for characterizing macromolecule-ligand interactions that has been used as a hit finding tool with significant success. Recent advances in MS and separation technology have positioned AS-MS to impact more areas of drug discovery. This manuscript provides a brief historical review of AS-MS and the recently developed technologies that have enabled AS-MS. The report also provides examples and references for how AS-MS has been used for high-throughput screening (HTS) to DNA-encoded library (DEL) screening hit confirmation, Direct-to-Biology, and natural product screens. The references for this work were collected from a broad range of sources, including Google Scholar, Scopus, review articles identified via Google Scholar, and the internal AI resource at AbbVie Inc. AS-MS is a unique biophysical binding assay that does not rely on labels and can specifically detect binders from large pools of potential ligands based on molecular weight. There is still significant room for growth in areas of impact that will be driven by decreases in separation time and a move toward equilibrium conditions during separation. Increased use for driving rapid structure-activity relationships (SAR) has potential to decrease project cycle times in lead identification and optimization.
The opioid crisis burdens the health care system and poses major research challenges. One approach toward safer analgesics involves targeting opioid receptors in areas of tissue injury/inflammation. Increased proton concentrations have been successfully exploited for developing novel ligands with limited side effects. This review investigates the impact of other inflammatory components (free radicals) on opioid receptors, focusing on redox-sensitive thiols and disulfides. This narrative review is based on a systematic literature search up to 3 March 2025. Of the identified 938 articles, 29 articles met the authors' inclusion criteria. Risk of bias was assessed according to NINDS recommendations, and the review followed PRISMA guidelines. As the included studies indicate that disulfides are an essential structural component, it is conceivable that free radicals affect opioid receptor activity. Current drug design has largely overlooked that function of G-protein coupled receptors (GPCRs) can differ between healthy and pathological microenvironments. The success of pH-dependent opioid ligands illustrates the therapeutic potential of exploiting such conditions. Redox changes may represent another regulatory component, but functional and in vivo evidence is still lacking. The consideration of microenvironmental factors may enable the development of safer, peripherally acting analgesics and refine GPCR-targeted drug design.
Deep generative models are reshaping de novo drug design by enabling creation of novel, property-optimized molecules beyond traditional chemical libraries. Advances in deep learning, molecular representation learning, and structure-aware modeling now enable algorithms to propose molecules that satisfy complex pharmacological constraints, accelerating hit identification. This review outlines recent advances in generative molecular design, including neural network-based frameworks, reinforcement learning systems, diffusion models, and language model-based transformers. The authors outline how each class generates and optimizes molecular structures and review generative AI's practical applications in drug discovery, illustrating translational progress. Current bottlenecks are critically analyzed alongside emerging solutions. This review is based on a systematic literature search conducted in Google Scholar and PubMed, covering studies published up to December 2025. Generative AI's greatest promise lies not in generating more molecules, but in generating better hypotheses, structures that are synthetically accessible, biologically plausible, optimized across potency, selectivity, and pharmacokinetics. The next phase will be led by multimodal foundation models capable of reasoning jointly about chemistry, protein structure, and cellular response, supported by automated synthesis and high-throughput experimentation. As these components are integrated, generative molecular design will guide lead optimization and reshape how new therapies are discovered and developed.
Uveal melanoma (UM) is the most common primary intraocular malignancy in adults, but primary tumor treatment carries a high risk of permanent vision loss and does not adequately prevent metastatic progression. In vitro UM models are needed to accurately represent human disease to support translation of laboratory research to the clinic. This review covers current and emerging in vitro UM models. A PubMed search used keywords 'uveal melanoma' and 'cell line,' 'spheroid,' 'organoid,' 'culture,' or 'in vitro' to identify cell lines, three-dimensional (3D) cultures (spheroids and organoids), and co-culture systems. Model successes and shortcomings are described, considering features that make models more or less representative of in vivo human UM. Insights are provided for consideration when selecting UM models for novel drug discovery. While traditional cell lines have provided an important foundation for UM research, emerging spheroid and patient-derived organoid models may more accurately represent in vivo tumor behavior and the tumor microenvironment. Pairing these 3D models with co-culture techniques could dramatically improve the representativeness of UM models. Researchers should consider testing promising therapeutics on a panel of models representing different UM subtypes, with particular attention to high-risk UM, such as those with BAP1 loss.
Efficient structure-based design requires robust protocols for generating protein-ligand complex structures to support iterative chemical optimization. However, developing reliable crystallization conditions suitable for drug discovery remains challenging, especially for novel targets and when working with diverse ligand classes. The review focuses on establishing robust crystallization workflows and providing solutions when standard methods prove inadequate for obtaining protein-ligand crystal structures. In addition to reviewing the literature for generic technical advances, the authors provide a comprehensive overview of project- and protein-specific approaches. To further substantiate their claims, the authors analyzed metadata from their proprietary structure collection, representing 20 years of crystallography supporting structure-based drug design. The authors provide two detailed examples showcasing rescue strategies in action. Crystal structures will remain fundamental to structure-based drug design moving forward. Successful crystallization demands adaptable, multi-faceted strategies that systematically explore diverse protein variants and crystallization conditions. Future progress depends on integrating AI tools for construct design with project insights and robust experimental workflows. Success ultimately hinges on synergy between innovative problem-solving approaches and deep expertise in navigating this rapidly evolving landscape.