Pharmaceuticals

Chronic wound infections are difficult to treat because they involve polymicrobial communities, biofilm-associated tolerance, impaired healing and repeated antimicrobial exposure. These factors promote persistence and antimicrobial resistance, while the development of new antibacterial agents remains costly and prone to failure. This review discusses a Digital Twin framework for supporting multitarget antibacterial discovery and formulation design in chronic wound infections. The proposed concept integrates microbiological and epidemiological data, bacterial target networks, structure-based screening of plant-derived metabolites, and formulation-dependent release prediction. Natural compounds are particularly relevant because many phytochemicals affect several bacterial processes, including membrane integrity, cell wall biosynthesis, DNA-related targets, quorum sensing, and biofilm formation. However, their use is often limited by poor solubility, low stability, and insufficient local exposure. Therefore, the framework also includes delivery-system engineering, such as electrospun fibres, polymeric nanoparticles, lipid-based carriers and controlled-release platforms. Machine learning models may then be used to link formulation descriptors with predicted release profiles and local antibacterial exposure. The practical output of the framework is a prioritised set of compound–target–carrier combinations. The proposed Digital Twin framework is conceptual and has not yet been experimentally validated; however, it may reduce empirical screening and support more rational selection of such combinations for chronic wound management.

Generative and predictive artificial intelligence (AI) is rapidly increasing the number of molecular, structural, cellular, and therapeutic hypotheses that can be explored computationally. Experimental capacity, however, has not expanded at a comparable rate. The result is an evidence bottleneck, in which biological hypotheses are generated faster than they can be experimentally tested, reproduced, and translated. This mismatch is particularly consequential for pharmaceutical discovery, where computational predictions alone cannot establish biological function, mechanism, or therapeutic value. As the number of computationally generated hypotheses grows, the central challenge shifts from candidate generation to the prioritization of hypotheses for experimental validation. This evidence bottleneck represents a defining constraint of AI-enabled biological discovery. We examine how different biological claims, including structural, interaction, functional, mechanistic, and therapeutic claims, require distinct types and levels of evidence. We also address the cost and context-dependence of experimental validation. Furthermore, we discuss how information-guided experimental design and iterative model–experiment feedback can improve the allocation of limited experimental capacity. We propose an evidence-centric framework for drug discovery that extends the evaluation of generative models beyond their ability to generate or rank hypotheses. The framework emphasizes experimental testability, mechanistic understanding, reproducibility, and translation into actionable biological and therapeutic knowledge.

Background/Objectives: Combining heterocyclic fragments provides a strategy for developing compounds for exploratory anticancer evaluation. This study aimed to synthesize and characterize five thiazolyl-methylthio-1,3,4-thiadiazolyl-quinazolin-4(3H)-one compounds (5a–e), assess their properties in silico, and investigate the selected ones for cytotoxicity and their effects on irradiation-associated DNA lesions. Methods: Structures were characterized by IR, HRMS, and ¹H- and ¹³C-NMR spectroscopy. Computational studies included physicochemical and ADMETox predictions, multi-platform target prediction, and molecular docking to A2A, Erβ, and CDK5. Compounds 5a and 5b were evaluated for cytotoxicity and cell-death profiles on one colorectal cancer (CRC) cell line (HT-29), two breast cancer (BC) cell lines (MCF-7 and MDA-MB-231), and one healthy fibroblast cell line (BJ), using cisplatin (Cis) as a reference. DNA lesions were assessed by the Comet assay in both breast cancer models before, immediately after, and 24 h after 4 Gy irradiation. Results: Compounds 5a–e were obtained in yields of 45.17–83.27%. Compounds 5a and 5b had lower 24-h IC₅₀ values than cisplatin in MCF-7 cells, but higher values in the other tested cell lines. Compound 5b showed higher selectivity indexes than both 5a and cisplatin across the cancer models, although its values of 1.14–1.34 indicated modest selectivity. Cell-death profiles differed between treatments, with larger necrotic fractions following 5a or 5b treatment in MCF-7 cells. Both compounds were associated with higher post-irradiation lesion scores than untreated irradiated controls. Scores increased further at 24 h in MCF-7 cells but decreased in MDA-MB-231 (triple-negative) cells. Pre-irradiation scores were also elevated in treated MCF-7 cells. Computational profiling identified comparatively more favorable predicted solubility for 5a and 5b. Cross-platform ADMET comparison consistently indicated low Caco-2 permeability, CYP3A4 inhibition, potential liver-injury risk, and high plasma–protein binding, while several absorption, transporter, and safety endpoints remained discordant. Target predictions showed weak cross-platform convergence. Docking provided the strongest, although moderate, structural support for A2A binding by 5a–d. Conclusions: The findings support further investigation of 5a and 5b, particularly their effects on irradiation-associated DNA lesions. Priorities include improving potency and selectivity, validating in silico predictions, and evaluating radiosensitizing potential through clonogenic survival studies.