Добавил:
kiopkiopkiop18@yandex.ru t.me/Prokururor I Вовсе не секретарь, но почту проверяю Опубликованный материал нарушает ваши авторские права? Сообщите нам.
Вуз: Предмет: Файл:
Ординатура / Хирургия / Библиотека им академика М.И. Перельмана / Книга_5606_Библиотеки_им_академика_М_И_Перельмана.pdf
Скачиваний:
0
Добавлен:
02.09.2026
Размер:
21 Мб
Скачать
s
n
C
y
3
g
n
is
s
g
QSA
12 Experimental Assays: Chemical Properties, Biochemical and... 357
nzymatic assay
rotein sequence
Expression/purificatio
Half-maximal inhibitory concentration (IC50)
rotein in solutio
rystallograph
D modelin
Detergent, BSA
Enzymatic inhibition
K
, K
cat,Kinact
I
ocking analys
simulation
Reversibility
hineLearnin
Fig. 12.2 Attaining enzymatic assays: Requirements and possibilities for experimental determina­tion in vitro and simulations in silico. A given protein sequence may be selected as a target for drug design and discovery and follow two distinct paths toward inhibitory activity determination. On one side, performing and achieving expression and purication may result in a stable protein in solution, which can follow crystallography studies. On the other end, a protein sequence can be modeled to a three-dimensional (3D) structure to perform computational simulations, such as docking, molecular dynamics (MD) simulations, and machine learning (ML) techniques, including comparative quan­titative structure-activity relationship (QSAR) analysis. Ultimately, various enzymatic assays can be performed to experimentally validate the target, considering inhibitory activity to determine the
values, reversibility, and K
IC
50
, K
cat
and Kiassays, including the possibility of disregarding
inact
unspecic aggregative compounds (e.g., detergent, or bovine serum albumin (BSA))
cellular staining for nonviable cells (e.g., trypan blue) [154], as well as the evaluation of the metabolic function with pH and redox indicators, or uorescence (e.g., resazurin and alamarBlue)[155] are broadly employed in research. Viability assays in vitro can be simple and less time-consuming methods that allow for evaluation in microplates, but may depend on the purpose of the assay as well as the availability of laboratory equipment and reagents [156, 157].
Overall, an assay should consider cost and sensitivity, while being rapid, safe, reliable, and reproducible without interfering with the test compounds [149, 157]. For instance, the crystal violet (CV) staining method (triphenylmethane dye (4-[(4-dimethyl amino phenyl)-phenyl-methyl]-N,Nimethyl-aniline) is an alter­native or complementary approach to indirect tests such as MTT and alamarBlue. Staining by CV depends on the viability of the cell monolayer in culture, that is, more CV is taken when there is a higher abundance of viable cells. This simple assay is a direct approach [158] useful for assessing the density of a cell monolayer while allowing for observation of morphological changes, and also allowing to storage the microplates for further analysis or future reference [159, 160]. Despite being simple and useful, CV and other assays may not offer substantial information for
358 M. Sá Magalhães Seram et al.
simulations in specic predictions, such as assessing synergism or antagonism effects [161], and may even not match predictions and comparisons between differ­ent methods for assessing specic molecules (e.g., nanoparticles) [162].
In this sense, some distinctions and specicities may be required for experimental validation, in which certain cell lineages or phenotype assays can be helpf ul. For instance, phenotypic assays, that is, experiments that assess alterations in the phe­notype of a given cell or organism [163], may display a predicted compound potency and selectivity to a proposed cell target while also being able to evaluate potential adverse effects in vitro [164]. This can be especially important in drug discovery approaches that aim for cytotoxic compounds, as in the search for anticancer drugs, where a phenotypic approach can account for the diversity and plasticity of molec­ular mechanisms and phenotypes in mutated cells, which could be indistinguishable from a general or indirect cell viability assay [165]. These phenotype assays can also benet from different toxicity predictions in silico [166] and even function as a validation method for assessing the adverse toxic effects of a compound in vitro [167], ultimately providing a more realistic disease model for translating simulations to biological determinations [168].
Moreover, considering the inuence of various targets in vitro, the use of chemical probes as reagents for improvi ng experimental validation [169] of targets modied by diseases can also benet cytotoxicity e valuation, facilitating the inves­tigation of target function and safety [170]. Chemogenomic and the use of chemical probes may also be helpful in drug predictions [171], as they are often more selective and differ in their chemical properties, allowing for comparing selectivity from small compounds evaluated for their cytotoxicity [170, 172]. Additionally, other reagents can be used for assessing target interactions and potential modications, such as intracellular signaling, ligand binding, and receptor interactions [173]. For example, by the improved luciferase assays with furimazine (NanoLuc) [174] and biolumi­nescence resonance energy transfer (BRET) in nanoBRET assays [175]. These assays exploit the energy donor from an enzyme (luciferase) to an acceptor (uorophore), followed by the oxidation of a given substrate, which results in a quantiable signal that translates interactions between molecules and proteins by proximity [173]. Herein, these assays can be used for experimental validation of cellular target simulations, as in the prediction of protein structures and stabilizing amino acids [174]. However, one should also account for potential interferences, such as inconsistent reagent concentrations and compounds acting as luciferase inhibitors, which could potentially impair expected results [175].
The interference of assay reagents in testing compounds accounts for one of the issues in translating predicted compounds to be assessed in cellular-based assays in vitro (e.g., promoting spurious results) [176]. For instance, compounds with antioxidant properties may generate false negative results by causing the reduction of substrate reagents, such as resazurin [176]. In addition, compounds can also be directly interferents in a uorescent assay, either quenching, that is, directly absorb­ing light that would be necessary for the
uorescent probe, or emitting light that would overlap the uorophore (i.e., presenting self-uorescence) [139]. This uo­rescence property was suggested for more than 5% (>70,000 tested compounds) of
12 Experimental Assays: Chemical Properties, Biochemical and... 359
the PubChem libraries in 2008 [177]. Similarly, ~12% (43,885 chemicals) of the bioactive compounds found in the NIH Molecular Libraries Small Molecule Repos­itory in 2012 inhibited the rey luci ferase to a certain level, displaying various luminescence changes [178]. Therefore, choosing a suitable method is pivotal to properly evaluate hits selected from simulations in drug discovery, as these behav­iors may also impact different experimental validation methods, including the enzymatic inhibition assays and impact corresponding targets (e.g., proteases) [179], as aforementioned.
A suitable uorescence-based method is also important for HTS platforms in drug discovery and toxicity screenings, where false readings could drive erroneous results for in vitro assessments and falsely corroborate previous negative or positive pre­dictions [139]. Regarding HTS, the Z-factor coefcient is a measurement that can help assure quality in a screening assessment [180], as it is a dimensionless statistical parameter that incorporates the means and errors associated with a given positive control and a negative reference, useful for comparison purposes as well as optimi­zation and validation [181]. In addition to assuring experimental determination quality, the success of simulations also needs to account for variations in the biological phenomena of cell cultures, the range and/or lter of absorbance, the availability of equipment, the ability to detect, and specicity of reagents used (e.g., uorophores) [157, 182]. Herein, in silico predictions have successfully identied particularly problematic chemical structures and substructures that could interfere in such biological assays, such as quinones capable of forming covalent adducts with thiols [183] and the presence of compounds that might exhibit autouorescence and interfere with uorophores [184].
However, pan assay interference compounds (PAINS), that is, chemical struc­tures that present a non-selective activity [185] or multiple behaviors that interfere with assays [186], are usually false positive hits from computational predictions. PAINS can interfere with various assays, such as cell viability and enzymatic inhi­bition, displaying nonspecic cytotoxicity and promiscuous target interference [138]. They may not be easily or properly predicted in simulations, thus requiring specic screening or ltering of substructural features to be identied and avoided [185]. Here MD-based protocols may help simulate the propensity of a given compound effect as interferent (e.g. , lipid membrane perturbation) [186], but assessing libraries in a large-scale would require formulated and appropriate lters to recognize different PAINS among thousands of screened compounds to exclude them from further analysis [138].
The natural product curcumin, for example, a PAINS that has displayed various biological and inhibitory activities [ 187], could be associated with intercalation in the lipid bilayer of mammalian cells [188], as well as changing the structure of bacterial membranes [189] or viral envelopes [190], thus being associated with a nonspecic toxicity [63]. This potential disturbance or intercalation of lipid mem­branes could be assessed by zeta potential assays in vitro, which can measure the electrokinetic potential of colloidal suspensions, that is, the behavior of a colloidal slipping movement under an electric eld (e.g., lipid bilayer) [191]. The zeta potential can be simulated in silico, for example, using nonequilibrium MD
360 M. Sá Magalhães Seram et al.
simulations, which can predict temperature and ion inuences and differences in diverse surfaces [192]. However, even though it may be accurately predicted [193], an observed decrease in cell viability may not always be positively correlated with zeta potential data, as the assay demonstrates the enhancement of membrane per­meabilities and not direct disruption [63, 194], as those observed for curcumin itself [195].
Moreover, compoundssolubility should also be considered for experimental validations, as in enzymatic and cell viability assays. Some simulations can be employed as qualitative and quantitative approaches aiming to predict solubility in water-based systems [196]. For example, a quantitative structure–property relation­ships (QSPR) model may allow for the prediction of solubility [136] providing the quantied numeric values with an uncertainty measured by a root mean square error (RMSE) of the model, helping in a direct comparison between two chemical structures [197]. In addition, similar properties can also be quantitatively predicted, which benet from structural analogs and common scaffolds, but can also face difculties when assessing large datasets and structurally diverse compo unds retrieved from diffe rent chemical libraries [136]. However, solubility issues such as saturation and precipitation may be troublesome for many in vitro assays, potentially promoting unwanted effects, such as variations in permeation [198]. These may result in false negatives or false positives even in high concentra­tions of dimethyl sulfoxide (DMSO) [137], which is usually a universal solvent option that can dissolve compounds presenting a wide range of physicochemical properties [199].
Physicochemical properties themselves may ultimately impact in PK or ADMET parameters. To ones advantage, these parameters can be simulated in compu tational predictions [18], including a qualitative classication of being active or inactive (e.g., hepatotoxicity) [200], and positive or negative (e.g., intestinal absorption) [201]. While there is no ideal PK parameter for any drug candidate, as it depends on each target, various tools are mostly freely available as integrated platforms on web servers aiming to accurately and comprehensively predict ADMET and/or PK properties of assessed c ompounds [202]. For instance, they can target specic receptors (e.g., Pred-hERG [203]) or a broader set of predictions in a single simu­lation (ADMETlab [204]), which may have been updated and/or upgraded (ADMETlab 2.0 [205]). However, experimentally validating ADMET and PK simulations from these web servers would require specic evaluation in vitro or in vivo (e.g., toxicity in specic cells or animal models), which could be limited to laboratory structure, expertise, and resources while not being cost-effective. In this sense, the major recommendation for these predictors is to be used in combination as consensus predictors to increase individual accuracy (usually >70%) and approxi­mate a comparative evaluation, potentially improving the success of drug candidates [18].
12 Experimental Assays: Chemical Properties, Biochemical and... 361

4 Antiviral Assays in Experimental Validation

The drug design, discovery, and development of new antiviral drugs are one of the main focuses in medicinal chemistry, especially when considering that various known and new viral diseases are yet to come, with the emergence and reemergence of different viruses [73]. In the search for new antivirals, various computational approaches (e.g., VS) have played a key role and were successful in o btaining lead candidates that later were licensed as antiviral drugs [206]. For instance, one can cite successful cases in the antiretroviral therapy of human immunodeciency virus (HIV), especially protease inhibitors [207], which beneted from CADD approaches in the initial design of lead candidates and led to the approval of drugs targeting proteases [208], such as of ritonavir [209] and saquinavir [210]. Other examples of success include oseltamivir and zanamivir as therapeutic options for inuenza virus [211, 212], the viral protease inhibitor boceprevir for hepatitis C virus (HCV) [213], and more recently the SARS-CoV-2 M diseases are considered a public health concern worldwide, and can progress from outbreaks to endemic, epidemic, and pandemic scenarios [214]. Still in 2019, the World Health Organization (WHO) declared different viruses as the top 10 global threats to be faced in the near future [215], months before the emergence of SARS­CoV-2 [216], yet unknown at the time as a novel coronavirus (2019-nCoV) [217].
Many viruses, including coronaviruses, are considered as threatening pathogens that can cause human diseases with high dissemination, morbidity, or mortality, such as the acquired immunodeciency syndrome (AIDS), COVID-19, Ebola, inuenza, and dengue and Zika fever [214, 218], some of which still do not have antiviral drugs as therapeutic options [219]. The urge of SARS-CoV-2 and the COVID-19 pan­demic, for example, promptly required collective efforts and resources to tackle the scenario, including strategies focusing in the discovery of new or repurposed antivirals [220, 221]. In support of this notion, between January 1981 and September 2019, 81 small molecules were approved by the Food and Drug Administration (FDA) as antiviral drugs, accounting for 44% of approvals aside from peptides, proteins, and vaccines [222]. As antivirals are usually designed to target a single microorganism, they can be direct-acting antiviral agents (DAA) used in a combined therapy (e.g., antiretroviral therapy against HIV), or aim for an induced cell response [223]. On the other hand, aiming for broad-spectrum activity usually does not display favorable outcomes due to the target specicity in different viruses [224, 225], which may result in a lower antiviral activity or high cytotoxicity [226]. Despite that, multi-virus or multi-target activity could be still predicted in simulations and employed to urgent scenarios, or in the preparedness for future emerging or reemerging viruses [221], as was the case of CADD approaches against different coronaviruses [227].
Initially, to validate multi-virus or multiple targets activity, one should consider a compound ability to permeate cell membranes and reach a given virus target (e.g., protease), or block viral entry in ho st cells, as viruses are obligate intracellular microorganisms [228]. Here, certain challenges are expected, such as a target
pro
inhibitor, ensitrelvir [29]. Most viral
362 M. Sá Magalhães Seram et al.
specicity and consequent lower cell or host toxicity, and the necessity of intracel­lular availability and efcacy [226, 229]. Aside from predicting a given compound permeability, one should also consider that a virus will have a cell, tissue, or host tropism, that is, the preference to infect a certain cellular type [230]. In addition, a cell must be permissive to the virus, that is, support the infection by having receptors that allow for viral attachment and entry, as well as allowing for intracellular replication, thus successfully resulting in viral progeny or subsequent infections [231]. This is important when dening which cell platform will be assessed in vitro, but also to design specic predictive models toward a certain cell, cell receptor, or protein of interest, such as models assessing inhibitors of the angiotensin converting enzyme 2 (ACE2) receptor for blocking the SARS-CoV-2 variants entry in mam­malian cells [232]. Furthermore, some viruses require a biosafety level 3 (BSL3) facility to be assessed in experimental validations, which can also be a limiting factor for the development of antiviral compounds [233], and potentially require an adaptation for BSL2 models in vitro [234] and in vivo [235].
Even if able to successfully move a computational simulation to an experimental validation, one should also account for fast replication or mutation rates of different viruses in vitro, which can impact the antiviral activity and result in a lead drug can­didate failure [236], especially for viruses that cause chronic infections, such as hepatitis B virus (HBV), HCV, HIV, and that might include drug resistance [73, 229,
237]. In addition, by requiring specic cell lineages and culture conditions add to the
reasons why the design and discovery of antivirals may have a slower rate [238], especially when compared to antibacterials or antifungals [73, 239]. Additionally, standard protocols for propagation, purication, and titration of viral strains may be needed [240, 241], as well as standardizing viral adsorption and incubation periods, and determining the multiplicity of infection (MOI), that is, the number of viral particles calculated to infect cells in culture. This number is crucial for some antiviral experiments, as the higher the MOI, the faster the cell viability reduces due to the cytopathic effects (CPE) caused by a higher number of viruses. In this sense, one-step and growth curves can contribute to evaluating the best MOI of work, while observing a more suitable log-phase in viral propagation to elaborate an optimal strategy of the study [240, 241].
Moreover, purication protocols may require specic centrifugation or ultracen­trifugation protocols before testing [242], all aiming to establish the best conditions to assess compounds against certain viruses in vitro and subsequently in vivo. These may be challenging to translate from computational simulations, either requiring individual approaches for each parameter to be tested (e.g., cell permeability) [243], or not being able to be predicted (e.g., purication conditions). Thus, even after predicting a given compound, experimental validati on should consider different antiviral assays (Fig. 12.3). For example: (i) pre-treatment, aiming to identify potential compound-induced cellular antiviral responses; (ii) virucidal activity, with direct action of compounds against viral particles before infection; and (iii) neutralization or reduction assays, assessing a direct effect of a given compound in a step of the virus replication cycle, such as viral entry, uncoating, replication, protein synthesis, morphogenesis/assembly, or release [244, 245].
12 Experimental Assays: Chemical Properties, Biochemical and... 363
Cellular-based assays
Cytotoxicity Antiviral
Successful approach
Fig. 12.3 Considerations for computational simulations translating into cellular-based assays (cytotoxicity and antiviral). For a given compound toxicity assessment against cell lineages in vitro, one should consider potential precipitation, as well as the interference of pan assay interference compounds (PAINS) in assays, before following absorption, distribution, metabolism, excretion, and toxicity (ADMET) or pharmacokinetics (PK) simulations (white box). Ultimately, compounds may be disregarded or follow in vivo studies. In parallel, computational simulations may support cellular or viral propagation conditions, acknowledging potential limitations regarding temperature and incubation periods in cell infection, or purication predictions of a viral parti­cle (light gray box). These may help in the prediction of an antiviral compound. Lastly, if a biosafety laboratory level is not a restriction, and a permissive cell lineage is available, cytopathic effects (CPE) can be validated in vitro and follow various assays for determining a compound antiviral activity, such as plaque neutralization or reduction assays (dark gray box). On the other end, chronic infection models and high mutation rates may impair experimental validation, while potential inhibitors may be inactive, or require specic viral target validation. Bold lines represent a successful experimental path, while regular lines represent issues or obstacles
Furthermore, various targets were considered and assessed for the design of potential antiviral candidates against SARS-CoV-2 proteins, including structural (e.g., spike [246, 247]), and non-structural associated with viral replication (e.g., viral proteases [79, 248] and RNA polymerase [249]). As for protein-pr otein inter­actions (PPI) from the viral attachment protein to host cell receptors, various predictive ML models [250, 251] and tools (e.g., DeepViral [252] and MP-VHPPI [253]) have been proposed, including DL models [254]. These can benet the design of target specic compounds as antiviral inhibitors, for example, in approaches that may benet from the availabil ity of chemical structures [255] and previous predic­tive ML models [254]. Among these possibilities, predictive or classication ML could help in training DL models from specic antiviral databases as sources of different chemical structures, such as experimentally determined peptides (AVPdb [256]) and those classied by ML models, such as AVP-IC
[257], a viral annotated
50
antiviral activity database from ChEMBL [258]. In addition, virus-targeted data­bases, such as the Inuenza Research Database [259] and the Coronavirus Antiviral Research Database (CoV-RDB) [260], could also be used for feeding models.
364 M. Sá Magalhães Seram et al.
Hereupon, with the large availability of SARS-CoV-2 experimental data from the search of potential antiviral drugs in the recent years, developing new focused databases with experimental validation data (e.g., CoronaDB-AI [261]) could also improve the success rate in the discovery of new lead candidates against other viruses [262, 263] and other viral targets [264266], such as those of HCV and HIV [267], and even open possibilities to other microorganisms [268], such as bacteria and fungi.
5 Determining Antibacterial and Antifungal Bioactive
Compounds In Vitro
In the eld of drug discovery, the development of new antimicrobials facing infectious diseases also consider, along with antivirals, antibacterial and antifungal compounds [269]. These have been known for over a century and have also inuenced the current knowledge for drug design and discovery [270]. The increas­ing demand for novel antimicrobials against the global spread of multidrug-resistant (MDR) pathogens, for example, may be faced by various sectors of research and funding, aiming for a bridge between academic, governmental, industrial, and pharmaceutical efforts that can translate expertise and data from computational and experimental studies into a successful drug candidate [271].
Regarding MDR bacteria, there is a lack of newly approve d antibiotics [272], contrasted with estimates that overcome more than $1.7 billion in increased healthcare costs per year related only to methicillin-resistant Staphylococcus aureus (MRSA) [273], which approximates the costs to develop a single new drug [1]. Fur­ther, back in 2019, AMR was considered one of the top 10 global threats to the future by WHO [215], which reected the critical current scenario of ~700,000 deaths yearly due to MDR infections [274], and estimates of over 10 million by 2050 [275]. For instance, deaths by MRSA infections surpassed those from HIV/AIDS and tuberculosis combined in the US [276], which is also considered a serious threat by the Centers of Disease Control and Prevention (CDC) [273], demanding a high priority for the search and development of new drugs [277].
On the other hand, fungal diseases reect the concern for primary or co-infections, such as those caused by the emer ging Candida auris, which has drawn attention since it was rst identied in 2009 and is rapidly rising to a global risk in healthcare clinical settings as an MDR pathogen with mortality rates ranging from 30% to 60% [278]. After being declared an urgent health threat by the CDC in 2019 [273], a warning on C. auris was issued in early 2023, summarizing clinical cases (present infection) increasing from 476 in 2019 to 1471 in 2021 and 2377 in 2022, in addition to screening cases (i.e., detected but not causing infection) that tripled in the US [279]. In the meantime, as the COVID-19 pandemic may have impacted invasive fungal infections in public health and antifungal drugs usage [280, 281], including COVID-19-associated C. auris outbreaks (mortality rates up to
12 Experimental Assays: Chemical Properties, Biochemical and... 365
83% [282]), MDR candidemia cases are spreading and increasing in hospital settings worldwide [283], with the ability to form multilayer biolms also resisting various niche skin conditions (e.g., desiccation) [284].
In this sense, computational methods may provide reliable and cost-effective strategies at different stages of drug design and discovery of new antibiotics [271], as in efforts for safer and more effective drug candidates needed for MDR infections [285]. For instance, predictive ML techniques may help in the screen ing of antibacterial and antifungal compounds, for example, feeding models for selecting natural products, semisynthetic derivatives, or synthetic compounds, as was the discovery of halicin, a potential antibacterial drug candidate discovered by a DL model (for more details on machine learning and neural networks, see Chap. 4) [286]. This compound, structurally different from many antibiotics, successfully supported a computational models ability to predict potential drug candidates based solely on other chemical stru ctures [286]. As mentioned for antivirals [73], predic­tive ML models can not only provide improvements for other potential drug candidates [60], such as antibacterial [270] and antifungal bioactive compounds [287], but also for the search for potential antimicrobial targets [288] and drug resistance [289]. For instance, ML methods can be applied in simulations to verify the accuracy of a designed model against experimentally evaluated peptides [290].
However, when considering the prediction ability of ML models to discover potential bioactive compounds, the true positive rate of a given model can be dependent on the availability of a varied set of compounds (e.g., small, large, similar, and diverse libraries) and their biological information (e.g., experimental determi­nation). For instance, Deep-AntiFP, a tool for prediction of antifungal peptides with deep neural networks, reached the highest accuracy of 94.23% when using the training dataset to validate the models, but achieved 91.02% and 89.08%, when assessing an alternative and an independent dataset, respectively. Despite an overall reasonable accuracy (~90%), even when increasing the number of iterations in the models, the predictive error would tend to remain stable, indicating the need for a more diverse dataset as input, thus consequently requiring more experimentally validated compounds to increase the models accuracy [291]. In addition, even novel approaches that pre-trained a predictor model achieved only comparable performance to other available computational methods [292], as in a consensus QSAR-screening approach [293], which also corroborated the need for a larger and more diverse set of experimentally determined compounds for increasing predictive accuracy.
This would stand in contrast to the discussion of nding an equilibrium between the cost of experimental testing and enough supporting evidence before selecting a target or compound to be tested in drug discovery approaches [294]. Building the most accurate model for each approach would be ideal, but one should consider the fact that it is improbable to generate or deploy models with 100% accuracy, and that adding compounds or data that do not belongwould not be benecial to a model. In addition, it is unlikely to have all the chemical or biological information necessary to achieve high accuracy models, even when using previously solved studies, which also depend on the amount of information available [295]. Hereupon, virtually
366 M. Sá Magalhães Seram et al.
screening compounds could still be the major approach to discovering new com­pounds, regardless of accuracy, as shown for halicin [286]. Thus, as aforementioned, the consensus approaches that combine VS with AI could increase success and reduce false positive hits predicted with single computational methods [ 206].
Identifying a compound predicted to have a given biological activity, such as antibacterial or antifungal, would require further characterization in vitro. For example, starting with a minimal inhibitory concentration (MIC) determination and further assessing a proposed mechanism, such as bactericidal activity with minimal bactericidal concentration (MBC) determination, might help progress to in vivo evaluation [296]. In this sense, initial screening assays, such as broth microdilution assays in 96-well microplates [297], may be an easy, fast, and low cost method [298] to be used in early stages for antimicrobial activity. Finding if a given compou nd is active or not may even require further investigation of potential mechanisms (e.g., association with a specic infection or microorganism), or focus solely on bioactive compound identication [ 299]. However, validation assays should also account for difculties in culturing a targeted microorganism, such as slow-growing mycobacteria or intracellular bacteria [300], as well as the existence of non-culturable microbes with pathogenic potential [301], and the possibility of some microorganisms undergoing a viable but non culturable state, as some Candida species [302].
Pretomanid, for example, a drug for the treatment of MDR Mycobacterium tuberculosis, was rst identied in 2000 from a series of 100 compounds assessed for their antimycobacterial activity [303], which were repurposed from an original investigation as radiosensitizers for chemotherapy [304]. Empirically or randomly testing compounds, despite being known as usually not cost-effective [305], can still be an alternative when it is difcult to sustain experimental studies based on simulations from computational tools (e.g., validating a target) [306] or when assessing a relatively small number of compounds from small chemical libraries in 96-well microplate assays [294](e.g., MIC assays). These could lead to the identi- cation of compounds with activity against various bacteria but also result in cellular cytotoxicity due to nonspecic mechanisms [307], such as in cell membrane distur­bance, structural changes, damage, and potential disruption [63].
However, with a large amount of biological information and experimental data available for some pathogens, such as M. tuberculosis and S. aureus [73], some methods can provide reasonable results for translating simulations into biological assays, despite the fact that they could still lack the ability to predict novel resistance mechanisms [308]. Chemical similarity methods could be of interest when simulat­ing novel antibacterial compounds based on existing antibiotic drugs, as shown in the identication of crizotinib as a potential repurposing antibacterial candidate [63
]. However, these and other methods focusing on repurposing candidates may be limited by nite experimental resources, that is, a nite amount of available licensed drugs [309]. Lastly, the alternative of searching for repurposing candidates also falls under the recommendation of employing consensus approaches that maximize the number of available options for testing, usually providing consistent