In the early stages of drug research, in vitro models are critical in establishing whether a drug candidate is a transporter substrate or inhibitor. In vitro screening of drug–transporter interactions helps to predict the susceptibility of DDI mediated by transporters in vivo. In vitro models provide information related to the kinetic parameters of substrates (Km and Vmax) and inhibitors (Ki, IC50, or Vmax) for analyzing potential drug interactions ^[1][60]. Methods for in vitro studies of renal transporter-mediated DDIs mainly include membrane-based assays, cell-based assays, and renal slice uptake assays.

The membrane vesicle transport assay was the first membrane-based assay widely used in the study of the ABC superfamily. Membrane vesicles are formed by separating the plasma membrane from the cell expressing the transporter protein and providing ideal conditions (pH, temperature, and cofactors) for the preparation of membranes containing inside-out vesicles, ATP binding sites, and substrate-binding sites toward the outer buffer. If a drug accumulates within the vesicles in an ATP-dependent and concentration-dependent manner, it indicates that the transporter under investigation is engaged in its transport. If, on the other hand, a medicine inhibits the accumulation of the probe substrate, it is deemed a transporter inhibitor ^[1][2][60,79].

The membrane vesicle transport assay can be easily adapted to a multi-well plate format, making it suitable for high-throughput analysis. In addition, the use of commercially available vesicles can eliminate the variability between laboratories and preparations as they are produced in large quantities and can be stored at either −80 °C or in liquid nitrogen while maintaining their activity. Despite its advantages, the vesicle assay has a significant restriction when employed for hydrophobic substrates, as it can lead to false-negative results due to non-specific binding and vesicle leaking. However, this limitation can be minimized by using a rational study design, such as consistent procedures, consideration of compound solubility, and an appropriate incubation time ^[3][80]. Deng, F et al. ^[4][81] tested 232 drugs using membrane vesicle transport assays and found that many of the drugs already on the market were inhibitors of BCRP.

The early stage of development frequently employs ATPase for assessing efflux transporter interactions and screening DDIs. The colorimetric approach can be used to measure inorganic phosphate from ATP hydrolysis and reflect the transport activity ^[1][5][60,82]. For example, Satoh, T. et al. ^[6][83] used a human P-gp membrane ATPase assay in order to research the impact of Kampo medicines on P-gp and the DDIs between Kampo drugs and western pharmaceuticals. The study found that the majority of Kampo medicines inhibited ATPase activity, suggesting that they might inhibit P-gp function.

As the ATPase assay does not analyze medicines using radioactivity or specialized equipment, it is appropriate for high-throughput screening and allows for the batch analysis of compounds that interact with ABC transporters. However, the ATPase assay has some limitations. Firstly, the assay cannot directly identify transporter substrates or inhibitors due to its indirect measurement of transport. Second, some substrates and inhibitors’ ATPase activities do not match their transport rates, which can lead to results that are either falsely positive or falsely negative. Furthermore, a larger concentration of substrate is required, etc. ^[5][82].

Primary cells are produced from intact tissues and are capable of expressing all transporter genes found in that tissue. They can be used to study drug metabolism, transport, and clinical drug interactions. Janneh, O. et al. ^[7][84] assessed DDIs at the level of drug transport using peripheral blood mononuclear cells (PBMCs). The researchers used flow cytometry to measure the expression of P-gp, MRP1, and BCRP in PBMCs. However, although P-gp, MRP1, and BCRP were detected in PBMCs, there was no observed correlation between their expression and drug accumulation. As a result, an interaction at the transporter level does not fully account for the high failure rates observed with drugs such as tenofovir, abacavir, and [3H]-lamivudine. Lash, L.H. et al. ^[8][85] discovered that primary cultures of human proximal tubular cells expressed a varied array of transporters for important classes of significant drugs and were useful for investigating drug transport and disposal, as well as assessing potential DDIs in the human kidneys.

When characterizing transporter activities, using primary cells offers various benefits. The primary cells have an intact cellular architecture, a functional membrane, cytoplasmic elements, and cotransporting ions, which can lead to more reliable results. In addition, the primary cells are derived from intact tissues and can express all transporter genes present in that tissue. However, several unique transporters are commonly not expressed in primary cells, and transport tests are often performed in stable cell lines or transfected cells expressing the special transporters ^[5][82].

There are numerous cell lines used to construct transfected cells. As transporter-transfected cells, MDCKII, LLC-PK1, Chinese hamster ovary (CHO), HEK293, and HeLa cells have been extensively used. The construction process includes the transfection of recombinant plasmids containing the target gene fragments into specific cells, screening with G418, the selection of monoclonal cells, and the verification of the successful construction of transfected cells by combining Western blot, PCR, immunofluorescence microscopy, the radiolabeled substrate uptake/exclusion rate, etc.

Many single-transfected cells, such as OAT1/3-HEK293 ^[9][86], OCT2-HEK293 ^[10][87], PEPT2-HELA ^[11][88], and BCRP/MDR1-MDCK ^[12][89], have been successfully established. In a study, ranitidine absorption and its inhibitory effects on other medicines were examined in HEK293 or CHO cells that had been stably transfected with OCT1, OCT2, or their allelic variations. The results showed that ranitidine had the potential to cause DDIs when coadministered with OCT1 substrates and that OCT1 genetic variations significantly affected ranitidine absorption ^[13][90]. In addition, the fundamental mechanism of renal DDIs can also be understood using double-transfected cell lines ^[1][60]. Although it is difficult to construct double-transfected cells, MDCK-OCT1/2-MATE1 ^[14][91] and MDCK-OATP4C1-P-gp ^[15][92] cell models have been successfully established. König et al. ^[14][91] constructed MDCK-OCT1-MATE1 and MDCK-OCT2-MATE1 double-transfected cell models and used the corresponding single-transporter transfected cells as controls to research the transport mechanisms of metformin and the methyl–phenyl–pyridine cation, and confirmed that OCT1 and OCT2 mediated the uptake transport of both drugs, while MATE1 mediated the efflux. George, B et al. ^[16][93] investigated MDCK cells by double-transfecting OCT2-MATE1 and discovered that 5-HT3 antagonist drugs had the potential to inhibit the renal secretion of cationic drugs by interfering with the function of OCT2 and/or MATE1. Müller et al. ^[17][94] used double-transfected MDCK-OCT2-MATE1 cells as a model to simulate the organic cation transport processes in proximal renal tubule cells and to investigate the significance of OCT2 and MATEs in the interaction between cimetidine and metformin. The above approach allows not only to examine the transporters that mediate the transport of each substrate, but also to study the interactions of multiple transporters.

It is generally believed that single-transfected cells usually lack endogenous uptake or efflux transporters and cannot mimic the complete mechanism of the transmembrane transport of drug molecules, while double-transfected cells can overcome this defect to a certain extent. However, in intact organs, the transport of certain compounds may be mediated by multiple transporters, and even double-transfected cells cannot fully predict the true process of in vivo transport, so some researchers have constructed triple-transfected cells. For example, Hirouchi et al. ^[18][95] constructed MRP2/MRP3/OATP1B1 and MRP2/MRP4/OATP1B1 triple-transfected cells for studying the vectorial transport of drugs through the transporter.

Using transfected cells can cause any type of cell to express the transporter of interest, even if the transporter is not expressed in that type of cell. Moreover, the expression of transporters in cells can be controlled to a certain level. However, it is worth noting that the transfection process may change the cellular environment in which the transporter is located, which may affect the true expression and function of the transporter.

Conventional cell culture does not adequately simulate the in vivo environment; therefore, its use as a model for relevant studies can be biased or affect the use of experimental data. Unlike conventional cell cultures, 3D cell cultures can better simulate the natural environment in which cells survive in an organism. Even simple spherical models can compensate for many of the shortcomings of monolayer cultures. These structures can provide oxygen, nutrition, and metabolites, resulting in different cell populations. However, there are some disadvantages to using 3D cells, such as the fact that 3D cell culture consumes more time and resources and is more costly. Additionally, the research process is more complex and requires more technical and equipment support ^[19][20][21][96,97,98].

There are many 3D cell culture technologies, which can be summarized into three categories: 3D hydrogels, cell aggregation, and culture scaffolds. As hydrogels have biophysical properties very similar to those of natural tissues, they can be used as efficient 3D cell culture matrices. Examples of hydrogel technologies are natural hydrogels, extracellular matrices, etc. Cell aggregation is the phenomenon of several cells coming together to form clusters. Cell aggregation is related to the physiological characteristics of the cells, especially the material structure of the cell wall, with technical examples such as suspension drop culture plates, low-adherence planes, etc. Culture scaffolds provide a physical support into which cells can enter to grow and perform their functions; technical examples are natural scaffolds (collagen and gelatin) and synthetic scaffolds (polystyrene, polyurethane, etc.) ^[22][99]. Vriend, J. et al. ^[23][100] developed a 3D microfluidic proximal renal tubular epithelial cell model and demonstrated the functional activity and drug interactions of P-gp and MRP2 and 4 by fluorescence-based transport assays. This model proved to be suitable for assessing the interaction of renal drugs with efflux transport proteins.

In the renal slice uptake model, the kidneys of anesthetized rats are removed and cut into slices of approximately 300 μm thickness. The slices are incubated in an oxygenated water bath, the drug to be tested is added, and the amount of drug taken up is measured. Renal slice uptake assays are often used to examine whether the drug is a substrate or inhibitor of OAT1/3 and OCT, to predict the renal transport characteristics of the compound, and to make predictions for later transfection cell experiments. The benefit of this model is that studies using human or animal kidney tissue can more closely resemble the in vivo environment.

For example, Yang, S. et al. examined the uptake of piperacillin and tazobactam using renal slices. The findings revealed that rOat1 and rOat3 mediated a beneficial interaction between the two drugs ^[24][101]. Zhang et al. conducted pharmacokinetic studies and uptake assays using rat renal slices and hOAT1/3-HEK293 cells to assess the potential DDI between bentysrepinine and entecavir. The study results led to the conclusion that it is safe to use bentysrepinine with entecavir in clinical practice ^[25][102]. Xu, Q. et al. utilized LC-MS/MS to determine the plasma and urine concentrations of entecavir after intravenous and oral administration in vivo, as well as assess entecavir uptake in kidney slices and transfected cells in vitro. The study findings demonstrated that OAT1 and OAT3 are DDI targets between entecavir and JBP485 (a dipeptide) ^[26][103]. The above studies on renal transporter-mediated DDIs all involved renal slice uptake assays, but they are not applicable to all renal transporter studies. Owing to the lack of a complete renal tubular network, renal slice uptake assays can only examine the uptake of drugs by renal secretory transporters, and not reabsorption transporters. In addition, it is worth noting that the cutting machine may damage the tissue structure, resulting in large errors in the experimental results.

In vitro models still have suboptimal predictive performance for clinically relevant drug interactions, despite being widely used, because these in vitro models do not allow for the investigation of the coordinated action of all transporters that occurs in real-life epithelial cells. Furthermore, they frequently overlook the effects of drug metabolites, which cannot form in vitro ^[27][16].

Rodents (rats and mice) and non-rodents (monkeys) are the two types of animals that are most frequently employed in transporter-mediated DDI research ^[28][104]. Rodents, such as Sprague–Dawley rats, Wistar rats, and wild-type mice, have been widely utilized as the first species for pharmacokinetic studies because they are affordable, have good reproductive performance, and are resistant to infectious diseases ^[1][60]. To investigate whether probenecid can alter the in vivo pharmacokinetics of biapenem, Li, W. et al. dissolved biapenem and probenecid in saline and injected it intravenously through Sprague–Dawley rats’ tail veins and then collected plasma and urine samples for study. The researchers concluded that renal tubular secretion mediated by OAT3 is a minor pathway for biapenem clearance. Biapenem would be safe to use in conjunction with other antibiotics and antiviral medicines in a clinical environment ^[29][105].

The use of animal experiments allows for a more comprehensive understanding of drug metabolism and transport in vivo. By comparing differences between species, the properties of transporters can be further investigated. However, the huge disparities between native animal models and humans in terms of transporters make it challenging to interpret the results. New preclinical models, such as gene knockout animal models, have been developed as a result of the growth of molecular biology and engineering tools.

Schinkel et al. ^[30][106] used gene editing technology to construct mdr1a (−/−) mice, which have no obvious physiological defects, but do not express P-gp. Subsequently, there have been increasing types of gene knockout animal models with improved commercial applications, and a series of gene knockout mice related to drug transporters have emerged, such as BCRP (−/−) mice, MRP2 (−/−) mice, MRP4 (−/−) mice, PEPT1 (−/−) mice and OCT1 (−/−) mice, and so on ^[5][82]. Gene knockout animal models have become one of the most important tools for studying transporter functions today. In one study, Breedveld, P. et al. ^[31][107] used BCRP (−/−) mice and wild-type mice to assess the mechanism of interaction of benzimidazole with MTX. The conclusion was as follows: the clinical interaction between MTX and benzimidazole may be explained by competition for BCRP. In addition to this, Kikuchi, R. et al. ^[32][108] used OCT1/OCT2 double-knockout mice to study the clinical drug interactions of veliparib with renal transporters and found that OAT1/3-, OCT2-, and MATE1/2K-mediated DDIs were the least likely.

Gene knockout animal models are of great significance for the evaluation of renal transporter-mediated DDIs, as well as drug disposition. The use of gene knockout animal models has the following benefits: ① they are closer to the physiological mechanisms of the human body; ② transporter function can be elucidated under physiological conditions; and ③ drug uptake and efflux by transporters can be studied without inhibitors. However, when using these models to understand transport functions, care should be taken, because the deletion of one transporter can often cause particular alterations in the expression pattern and function of other transporters, as well as alter the physiology of animals. For example, in MRP2-deficient rats, the expression of the MRP3 protein was greatly activated ^[33][109]. In addition, the limited variety, high cost, and interspecies differences in gene knockout animal models make it difficult to apply this technology for the high-throughput screening of transporter substrates and inhibitors ^[34][110].

PET is a non-invasive imaging technique that can study the distribution of radiolabeled medicines in various organs and tissues. Moreover, it allows for the real-time monitoring of drug metabolism and transport processes. Therefore, PET is the preferred method for quantitatively evaluating transporter-mediated DDIs at the tissue level. PET has been applied in both preclinical animal models and human subjects to evaluate the impacts of transporter-mediated DDIs on drug disposition across various organ systems, including the brain, liver, and kidneys ^[35][111]. For example, Hernandez-Lozano, I. et al. ^[36][112] evaluated the impact of different drugs, which may result in transporter-mediated DDIs, on the tissue distribution and excretion of ciprofloxacin. Additionally, they used PET imaging-based pharmacokinetic (PK) analysis. The outcomes demonstrated that DDIs can occur due to the inhibition of renal transporters by concomitant drugs, resulting in decreased urinary excretion and increased blood and organ exposure to ciprofloxacin. Another study ^[37][113] conducted in rats and pigs used ¹¹C-metformin PET to evaluate the function of OCT transporters in the kidneys and liver. The studies both emphasized the advantages of PET imaging-based PK analysis to evaluate transporter-mediated DDIs at a whole-body level. However, it is worth noting that PET technology also has some disadvantages, such as the fact that it requires the use of radioisotope-labeled drugs and has radiological safety issues, as well as expensive equipment, requiring a high level of technical support and operators.

In vivo PET imaging can accurately quantify the impact of transporter-mediated DDIs on the distribution of radiolabeled drugs in various organs and tissues. This novel method shows tremendous potential for elucidating the role of transporters in drug distribution and may prove helpful in the creation of new drugs. Continued research is being conducted to identify new PET radiotracers that target specific transporters, which is expected to expand the range of applications for PET in transporter studies ^[35][38][111,114].

Although a lot of results have been obtained from in vitro studies or animal experiments in the study of DDIs, it is often difficult to provide clinical guidance or application. The reason is that many results are conflicting, and the interactions that occur in vitro or in animal bodies may not necessarily occur in the human body ^[39][40][115,116]. Therefore, under the ethical guidelines of medical research involving human participants, research on healthy volunteers has gradually been carried out. For example, Arun et al. ^[41][117] evaluated the potential DDIs between sitagliptin and gemfibrozil in a study of 12 healthy Indian male volunteers. The researchers concluded that gemfibrozil significantly elevated the AUC0-∞ of sitagliptin, possibly by inhibiting hOAT3 transporters at renal tubules. In a study conducted by Morrissey, K. M. et al. ^[42][118], the impact of nizatidine on the pharmacokinetics and pharmacodynamics of metformin was investigated in 12 healthy volunteers. The results showed that nizatidine had no effect on metformin systemic concentrations or CLr, indicating that specific MATE2K inhibition may not be enough to generate renal DDIs with metformin.

This model can be used to directly understand the metabolism and operation of drugs and their metabolites in the human body. However, it is essential to note that various factors, such as gender, age, social habits, disease conditions, and genetic predisposition, among others, play a major role in DDIs, resulting in great interindividual variability. To take these factors into account, future research studies should involve large patient populations ^[41][117].

The IVIVE method essentially uses preclinical in vitro and in vivo data to find correlations and then conducts corresponding in vitro studies in humans based on this correlation in order to predict the CLr in the human body. Before human dosing or prioritizing specific types of clinical DDI studies, in vitro data need to be integrated, understood, and translated to support risk assessment. To date, there have been few examples of IVIVE involving renal transporters for DDIs. Shen, H et al. ^[43][119] discovered that the cynomolgus monkey may be useful in the support of IVIVEs that involve the inhibition of renal MATEs and OCT2. In addition, IVIVEs adapted from cynomolgus monkeys may provide insight into the risk of DDI in humans.

With the advancement of IVIVE technology, its applications are expanding rapidly, extending from the original animal species to the human species. IVIVE is seen as a potential method for forecasting transporter-mediated drug CLr or DDIs in humans. It can reduce the need for people to experiment on animals and reduce the burden caused by animal ethical issues. However, transporter-based IVIVE in humans is still in an early stage due to the following reasons: (1) Without specific substrates or inhibitors, it is hard to identify the contribution of different transporters. The involvement of numerous transporters with overlapping inhibitors may undervalue or overvalue the contribution of transporters in drug disposition. (2) Some transporters may have multiple drug-binding sites, which can further complicate predictions. (3) There is still limited knowledge on species differences in drug transporters, which also adds to the complexity of IVIVE. (4) It needs a lot of data. In a word, caution should be taken when extrapolating in vitro data to an in vivo setting ^[44][9].

Ex vivo perfusion (EVP) models, such as ex vivo perfused lung, intestine, brain, liver, and kidneys, are commonly employed in drug transport research. The process of perfusion may require complex equipment and technical support. These models allow for a more physiological environment to determine transporter functionality compared with in vitro experiments. Additionally, EVP models avoid the potential confounding effects of other organs on drug disposal that may occur in in vivo studies. EVP models examine the absorption or efflux role of transporters on drugs by adding selective inhibitors of transporters to the perfusate and comparing the differences in drug levels in the perfusate, tissues, organs, or plasma. By comparing the results of single- or multiple-drug combination perfusion, the competition of the same transporter by the combined drug can be studied ^[45][46][120,121].

The EVKP model can be used to study the metabolism and excretion of the kidneys. For example, Posma, R. A. et al. ^[47][122] investigated whether adding metformin prior to or during ex vivo isolated normothermic machine perfusion (NMP) of pig and rat kidneys affected its elimination. The metformin CLr was significantly greater than creatinine CLr, confirming metformin production during the ex vivo NMP of both rat and porcine kidneys. In addition to elucidating the CLr mechanism, the EVKP model can be used to predict DDIs. For example, Hori, R. et al. ^[48][123] utilized an isolated perfused rat kidney model to study the renal tubular secretion mechanism of digoxin and its interaction with quinidine or verapamil. Their findings suggested that digoxin is a substrate transported by P-gp and that P-gp inhibition causes clinically significant interactions with quinidine and verapamil.

Endogenous biomarkers, which are generally physiological substrates of drug-metabolizing enzymes and transporters, have recently emerged as effective tools for advancing the risk assessment of DDIs ^[49][124]. A number of biomarkers predicting renal transporter activity have been identified, including 6β hydroxycortisol (6β-OHF), taurine and glycochenodeoxycholate sulfate (GCDCA-S), thiamine, N-methylnicotinamide (NMN), creatinine, N1-methyladenosine (m¹A), and so on.

6β-OHF is a prominent endogenous in vivo probe that is commonly used to evaluate the inhibition of OAT3. It is synthesized by hepatic CYP3A4 and eliminated from the body through urine ^[50][125]. Imamura et al. used transporter-expressing cell lines to determine that 6β-OHF is a substrate of MATE1, MATE2-K, and OAT3 ^[51][126]. In another study, Imamura et al. used in vivo inhibitors, probenecid and pyrimethamine, to selectively inhibit OAT3 and MATEs, respectively, and investigate the role of OAT3 and MATEs in the urinary excretion of 6β-OHF in humans. The in vivo and in vitro results indicated that OAT3 significantly contributed to the urinary excretion of 6β-OHF and that 6β-OHF can be utilized to evaluate OAT3-mediated drug interactions in humans ^[52][127].

Taurine is an amino acid that can be acquired from the diet or be produced in the body, whereas GCDCA-S is a significant sulfated bile acid conjugate present in plasma and urine. Taurine and GCDCA-S have been suggested as endogenous probes for assessing the inhibition of OAT1 and OAT3, respectively ^[53][128]. Tsuruya et al. ^[54][129] investigated the effect of probenecid on the alterations in endogenous chemicals in plasma and urine samples using metabolomics analysis and found that taurine and GCDCA-S can be utilized as probes to assess pharmacokinetic DDIs involving OAT1 and OAT3, respectively, in humans.

Thiamine ^[55][130], NMN ^[56][57][131,132], and creatinine ^[58][133] have been shown in previous studies to be potential endogenous biomarkers of OCT2 and MATE1/2K in the kidneys. Among them, creatinine is commonly utilized as a biomarker for renal function. Several compounds were positively predicted to result in OCT2/MATEs-mediated DDIs based on in vitro–in vivo correlation studies between the terms of inhibition of OCT2 and MATE1/2K and clinically observed changes in serum creatinine or creatinine CLr. However, the changes in serum creatinine associated with OCT2/MATEs DDIs are frequently insufficient to support its use as a biomarker ^[59][134].

In recent studies, m¹A has been incorporated as a novel biomarker for OCT2 and MATE1/2-K. To extend the understanding of endogenous probes for OCT2/MATEs, Miyake, T. et al. ^[60][135] first investigated novel endogenous substrates of OCT2 by the metabolomic analysis of plasma and urine samples from wild-type and OCT1/2 double-knockout mice. After verifying the transport of the candidate compound m¹A by OCT2/MATEs, the researchers evaluated its utility as an alternative probe for clinical DDI studies in animals and humans. The results showed that m¹A could be an alternative probe for evaluating DDIs involving OCT2 and MATE1/2-K. In addition, Miyake, T. et al. ^[49][124] also demonstrated m¹A as a superior OCT2 and MATE1/2-K biomarker through a series of experiments in another study.

The use of endogenous biomarkers has been expanded to complex DDIs with several possible interaction sites. From these, DDI risks involving multiple drug transporters can be evaluated in the same projects. The plasma or urinary levels of these biomarkers can enable the monitoring of drug transporter activities without the need to conduct a clinical DDI study. While biomarkers are effective, for some drugs, the production of biomarkers may not be obvious enough. Additionally, because of the overlap in substrate specificity for individual transporters, it is unlikely that specific biomarkers will be identified for each transporter of interest ^[61][62][136,137].

In silico models have been used to predict 计算机模型已被用于预测DDIs, as it is impossible to test all possible drug combinations through experiments. Currently, the in silico models for predicting DDIs mediated by renal transporters mainly include machine learning methods (MLMs) and physiologically based pharmacokinetic (PBPK) models. The following is a discussion of these two models.，因为不可能通过实验测试所有可能的药物组合。目前，用于预测肾脏转运蛋白介导的DDI的计算机模型主要包括机器学习方法（MLM）和基于生理学的药代动力学（PBPK）模型。下面对这两种模式进行讨论。

MLMs are a type of ligand-based computational method used for classification interactions with target proteins. Examples of 是一种基于配体的计算方法，用于与靶蛋白的分类相互作用。MLMs include random forest (RF), support vector machine (SVM), recursive partitioning (RP), k-nearest neighbor (K-NN), and Bayesian models ^[63][64]. Recently, 的例子包括随机森林（RF），支持向量机（SVM），递归分区（RP），k-最近邻（K-NN）和贝叶斯模型[138，139]。最近，以贝叶斯模型为代表的MLMs represented by Bayesian models have been used to predict renal transporter-mediated 已被用于预测肾脏转运蛋白介导的DDIs. 。Sandoval, P. J. et al. ^[65] examined the inhibitory efficacy of ，P.J.等人[140]检查了400 or more compounds against the 种或更多化合物对OCT2-mediated uptake of six structurally different substrates. 介导的六种结构不同底物的摄取的抑制功效。Discovery Studio version 4.1 (版本 4.1（Biovia, San Diego, CA, USA) was used to generate and validate Laplacian-corrected naive Bayesian classifier models. The same threshold was used (50% inhibition or higher), as well as the same method of 5-fold cross-validation and receiver operating characteristic (ROC) calculation. Testing data sets containing 80 compounds were collated to assess the predictive capacity of training data and generate statistical results. These datasets have been used for the development of Bayesian machine learning analysis and prediction algorithms. In early preclinical drug discovery, it is possible to use the virtual screening of large libraries of novel structures to identify potential OCT2 interactions. This approach can be a cost-effective way to eliminate compounds that may pose a problem. The above-mentioned study demonstrated how substrate-specific OCT2 data were used to generate predictive computational models. In another study by the same experimental group, ，加利福尼亚州圣地亚哥，美国）用于生成和验证拉普拉斯校正的朴素贝叶斯分类器模型。使用相同的阈值（50%抑制或更高），以及相同的5倍交叉验证和受试者工作特征（ROC）计算方法。整理了包含80种化合物的测试数据集，以评估训练数据的预测能力并生成统计结果。这些数据集已用于开发贝叶斯机器学习分析和预测算法。在早期临床前药物发现中，可以使用对新结构的大型文库进行虚拟筛选来识别潜在的OCT2相互作用。这种方法可以成为消除可能造成问题的化合物的一种经济有效的方法。上述研究展示了如何使用底物特异性OCT2数据来生成预测计算模型。在同一实验小组的另一项研究中，Martinez-Guerrero, L. et al. ^[66] used a corresponding array of Bayesian models to generate predictions for each compound.，L.等人[141]使用相应的贝叶斯模型数组来生成每种化合物的预测。

In summary, Bayesian models have the advantage of being fast and interpretable by defining toxicological vectors in fingerprint features. In early drug discovery, MLMs can quickly predict the substrate or inhibitory properties of candidate drugs. These methods have the potential to help identify 总之，贝叶斯模型具有快速且可解释的优点，可以通过定义指纹特征中的毒理学载体。在早期药物发现中，传销可以快速预测候选药物的底物或抑制特性。这些方法有可能帮助识别现有药物中的DDI perpetrators among existing drugs and guide further experimental validation ^[40][67]. MLMs have the following advantages: handling complex data, automated analysis, and strong predictive ability. However, the need for large amounts of data is a disadvantage of MLMs.肇事者，并指导进一步的实验验证[116，142]。传销具有以下优点：处理复杂数据、自动化分析、预测能力强。然而，对大量数据的需求是传销的缺点。

PBPK models are mathematical models based on anatomy, biochemistry, and physicochemical principles. 模型是基于解剖学，生物化学和物理化学原理的数学模型。PBPK models connect a series of physiologically meaningful compartments or organs through the circulation system, describing the ADME process of drugs in the body. Currently, PBPK models have had important impacts on drug development and post-marketing stages. The latest FDA guidance on drug interactions recommends the use of PBPK modeling approaches to predict potential drug interactions. PBPK models can predict DDIs caused by different conditions, different individuals, and different mechanisms, and thus guide the development of further experiments or substitute some DDI clinical trials ^[68][69][70]. 模型通过循环系统连接一系列具有生理意义的隔室或器官，描述了体内药物的ADME过程。目前，PBPK模型对药物开发和上市后阶段产生了重要影响。FDA关于药物相互作用的最新指南建议使用PBPK建模方法来预测潜在的药物相互作用。PBPK模型可以预测由不同条件、不同个体和不同机制引起的DDI，从而指导进一步实验的发展或替代一些DDI临床试验[143，144，145]。PBPK models can be constructed using mathematical computing software, and most investigations into transporter 模型可以使用数学计算软件构建，并且大多数对转运体DDIs have been carried out using PKPB software, such as 的研究都是使用PKPB软件进行的，例如Symcyp (Certara, Sheffield, UK), （塞塔拉，谢菲尔德，英国），GastroPlus (simulation plus, PA, Lancaster, CA, USA), （模拟加，PA，兰开斯特，加利福尼亚州，美国），MATLAB (（MathWorks, ，Natick, MA, USA), Pk，马萨诸塞州，美国），PK-Simand MoBi (Bayer Technology Services, Leverkusen, Germany) ^[71]（拜耳技术服务，勒沃库森，德国）[146]. The following disadvantages need to be overcome when using 使用PBPK models: ① the complexity of parameter acquisition and calculation and ② the interpretation of the meaning of the results being relatively complex.模型时需要克服以下缺点：（<>）参数采集和计算的复杂性，以及（<>）对结果含义的解释相对复杂。

To quant为了使用体外抑制常数（Kitatively forecast DDI between cimetidine and metformin using in vitro inhibition constant (Ki) values, researchers created a new PBPK model of metformin. With in vitro Ki values, this model successfully reproduced the DDI between cimetidine and metformin. They concluded that the interaction between cimetidine and metformin is probably caused by the inhibition of MATEs by cimetidine, and not by OCT2 inhibition ^[72]. ）值定量预测西咪替丁和二甲双胍之间的DDI，研究人员创建了一个新的二甲双胍PBPK模型。利用体外Ki值，该模型成功再现了西咪替丁和二甲双胍之间的DDI。他们得出结论，西咪替丁和二甲双胍之间的相互作用可能是由西咪替丁对MATEs的抑制引起的，而不是由OCT2抑制引起的[147]。Asaumi， R. et al. ^[73][148] developed a rifampicin 开发了一种利福平PBPK model to forecast P-gp-mediated DDIs. This PBPK model accurately predicted P-gp-mediated DDIs with talinolol, digoxin, and quinidine in a variety of situations with varying dosages, durations, timings, and delivery methods. These results show that their rifampicin PBPK model may be used to predict DDIs with different P-gp substrates and investigate the number of DDIs in the intestine, liver, and kidneys.模型来预测P-gp介导的DDI。该PBPK模型准确预测了P-gp介导的DDI与他林洛尔，地高辛和奎尼丁在各种情况下，具有不同的剂量，持续时间，时间和给药方法。这些结果表明，他们的利福平PBPK模型可用于预测具有不同P-gp底物的DDI，并研究肠，肝和肾中DDI的数量。