You are accessing a machine-readable page. In order to be human-readable, please install an RSS reader.
ContinueCancelclearAll articles published by MDPI are made immediately available worldwide under an open access license. No special permission is required to reuse all or part of the article published by MDPI, including figures and tables. For articles published under an open access Creative Common CC BY license, any part of the article may be reused without permission provided that the original article is clearly cited. For more information, please refer tohttps://www.mdpi.com/openaccess.
Feature papers represent the most advanced research with significant potential for high impact in the field. A Feature Paper should be a substantial original Article that involves several techniques or approaches, provides an outlook for future research directions and describes possible research applications.
Feature papers are submitted upon individual invitation or recommendation by the scientific editors and must receive positive feedback from the reviewers.
Editor’s Choice articles are based on recommendations by the scientific editors of MDPI journals from around the world. Editors select a small number of articles recently published in the journal that they believe will be particularly interesting to readers, or important in the respective research area. The aim is to provide a snapshot of some of the most exciting work published in the various research areas of the journal.
Original Submission Date Received: .
JournalsActive JournalsFind a JournalJournal ProposalProceedings Series Topics InformationFor AuthorsFor ReviewersFor EditorsFor LibrariansFor PublishersFor SocietiesFor Conference OrganizersOpen Access PolicyInstitutional Open Access ProgramSpecial Issues GuidelinesEditorial ProcessResearch and Publication EthicsArticle Processing ChargesAwardsTestimonialsEditing Services InitiativesSciforumMDPI BooksPreprints.orgScilitSciProfilesEncyclopediaJAMSProceedings Series AboutOverviewContactCareers News PressBlogSign In / Sign UpSubmit JournalsPharmaceuticsVolume 15Issue 410.3390/pharmaceutics15041260
Submit to this JournalReview for this JournalPropose a Special Issue►▼Article MenuArticle MenuAcademic Editors
Nenad D. Filipović
Zoran MarkovicSubscribe SciFeedRecommended ArticlesRelated Info LinksPubMed/MedlineGoogle ScholarMore by Authors Linkson DOAJTran, T. Tuyet VanTayara, H.Chong, K. Toon Google ScholarTran, T. Tuyet VanTayara, H.Chong, K. Toon PubMedTran, T. Tuyet VanTayara, H.Chong, K. To/ajax/scifeed/subscribeArticle ViewsCitations-Table of ContentsAltmetricshareShareannouncementHelpformat_quoteCitequestion_answerDiscuss in SciProfilesthumb_up...Endorsetextsms...CommentNeed Help?SupportFind support for a specific problem in the support section of our website.
Get SupportFeedbackPlease let us know what you think of our products and services.
Give FeedbackInformationVisit our dedicated information section to learn more about MDPI.
Get InformationclearJSmol Viewerclearfirst_page Download PDFsettingsOrder Article ReprintsFont Type:ArialGeorgiaVerdanaFont Size:AaAaAaLine Spacing:Column Width:Background:Open AccessReviewArtificial Intelligence in Drug Metabolism and Excretion Prediction: Recent Advances, Challenges, and Future PerspectivesbyThi Tuyet Van Tran
Thi Tuyet Van TranSciProfilesScilitPreprints.orgGoogle Scholar 1,2,3
, Hilal Tayara
Hilal TayaraSciProfilesScilitPreprints.orgGoogle Scholar 4,* and Kil To Chong
Kil To ChongSciProfilesScilitPreprints.orgGoogle Scholar 5,*1Department of Electronics and Information Engineering, Jeonbuk National University, Jeonju 54896, Republic of Korea2Faculty of Information Technology, An Giang University, Long Xuyen 880000, Vietnam3Vietnam National University—Ho Chi Minh City, Ho Chi Minh 700000, Vietnam4School of International Engineering and Science, Jeonbuk National University, Jeonju 54896, Republic of Korea5Advances Electronics and Information Research Center, Jeonbuk National University, Jeonju 54896, Republic of Korea*Authors to whom correspondence should be addressed. Pharmaceutics 2023, 15(4), 1260; https://doi.org/10.3390/pharmaceutics15041260Submission received: 8 March 2023/Revised: 7 April 2023/Accepted: 14 April 2023/Published: 17 April 2023(This article belongs to the Special Issue In Silico Drug Testing and Optimization, Coupling Physical-Based Modeling and Machine Learning)Download keyboard_arrow_downDownload PDFDownload PDF with CoverDownload XMLDownload EpubBrowse Figures Versions Notes Abstract:Drug metabolism and excretion play crucial roles in determining the efficacy and safety of drug candidates, and predicting these processes is an essential part of drug discovery and development. In recent years, artificial intelligence (AI) has emerged as a powerful tool for predicting drug metabolism and excretion, offering the potential to speed up drug development and improve clinical success rates. This review highlights recent advances in AI-based drug metabolism and excretion prediction, including deep learning and machine learning algorithms. We provide a list of public data sources and free prediction tools for the research community. We also discuss the challenges associated with the development of AI models for drug metabolism and excretion prediction and explore future perspectives in the field. We hope this will be a helpful resource for anyone who is researching in silico drug metabolism, excretion, and pharmacokinetic properties.Keywords: drug discovery; drug metabolism; drug excretion; artificial intelligence; machine learning; deep learning; in silico method; web servers 1. IntroductionMetabolism and excretion are two important processes in pharmacokinetics. Figure 1 shows an overview of drug metabolism and excretion [1,2]. Metabolism is the biological transformation by which most drugs undergo a change in their chemical structure in the body to produce the expected therapeutic effects of a certain drug and be more easily eliminated from the body [3]. Drug excretion refers to the elimination of drugs or their metabolites from the body [4]. Drug metabolism can yield metabolites that differ greatly from the original drug’s physical and pharmacological characteristics [5]. The rate of metabolism dictates the length and strength of a drug’s pharmacologic effect. Drug metabolism also plays a role in multidrug resistance in infectious illnesses and cancer chemotherapy, and the effects of certain medications as inhibitors or substrates of enzymes involved in xenobiotic metabolism are frequent causes of adverse drug interactions [6]. Drug metabolism affects drug efficacy and toxicity in humans and laboratory animals. Metabolism is also responsible for the clearance of more than 70% of clinical medicines [5,7], so it has been extensively researched as part of drug research and development (R&D) efforts. Both metabolism and excretion are tightly regulated by the body to maintain homeostasis and ensure that harmful substances are eliminated from the body. Disruptions to these processes can lead to the accumulation of toxic substances, which can cause many health problems, such as kidney and liver damage, metabolic disorders, and drug toxicity [8]. Drug metabolism and excretion play critical roles in the pharmacokinetics of drugs and have important implications for the R&D of new drugs, as well as for the safe and effective use of existing drugs. By understanding and predicting drug metabolism and excretion, researchers can screen unwanted drug candidates and design new drugs with improved pharmacokinetics, reduced toxicity, and increased efficacy.Predicting drug metabolism and excretion by in vitro and in vivo research is one strategy. These experimental assessments of metabolism and excretion are typically time- and money-consuming. For instance, testing a CYP inhibition from a non-good laboratory practice costs about USD 1000 and takes one week [9]. Given the high expense of conventional drug R&D, numerous computational algorithms for predicting the metabolism and excretion of therapeutic candidates have been developed, allowing for the screening of a large number of chemical compounds and subsequently finding a small number of viable candidates [10]. Especially, in silico approaches are increasingly being used to predict drug metabolism and excretion, and are widely regarded as the best “fail early and fail cheap strategy”, allowing for lower costs, time savings, and thus lower attrition rates in the late stages of drug development.Artificial intelligence (AI) can now be employed across the entire process of developing new medicines [11]. AI methods are also increasingly being used in the field of drug metabolism and excretion to predict the potential of drugs to be metabolized and excreted by the body. The use of AI allows for the rapid screening of vast libraries of compounds, yielding useful insights into the compounds’ potential metabolism and excretion. Predictions made using AI techniques may be more accurate than those made using more conventional approaches since they can be trained using enormous amounts of experimental data. Predicting the likelihood of metabolic and excretory interactions between numerous drugs at once is a strength of AI systems that can aid in drug discovery. AI techniques can provide useful information on the potential for metabolism and excretion, reducing the time and money needed to conduct in vitro and in vivo experiments. Moreover, the use of AI techniques in drug R&D has the potential to increase both the safety and effectiveness of drugs by creating new compounds with enhanced metabolism and excretion. In this review, we have summarized the background of drug metabolism and excretion and highlighted the key properties of these processes. We took a deep dive into the most recent developments in the use of AI for medication metabolism and excretion prediction. We also provide the research community with a directory of publicly available resources for predicting metabolism and excretion. Research in this area has a number of obstacles, yet there is also promising future growth. We hope this review will be of interest to researchers working to enhance and develop several high-precision prediction models for drug metabolism and excretion. 2. Evaluation MetricsEvaluating the performance of AI methods is critical for measuring a method’s effectiveness and fairly comparing the score of various models [12]. In this review, we present the following evaluation metrics: coefficient of determination (R2), root mean squared error (RMSE), specificity (SP), sensitivity (SE), Matthew’s correlation coefficient (MCC), precision, recall, F1 score, accuracy (ACC), Jaccard score, and area under the receiver operating characteristic curve (AUC). The formulas are as follows: R 2=1− ∑ y i− y^2∑ y i− y-2 Range 0,1RMSE=1 N ∑i=1N y i− y ^ 2SE= TPTP+FN Range 0, 1SP= TNTN+FP Range 0, 1MCC= TP * TN −FP * FN TP+FN * TP+FP * TN+FN * TN+FP Range 0, 1ACC= TP+TNTP+TN+FP+FN Range 0, 1 Jaccard= TPTP+FP+FN Range 0, 1 Precision= TPTP+FP Range 0, 1Recall= TPTP+FN Range 0, 1F1 score= 2× Precision ×RecallPrecision+Recall Range 0, 1 AUC= Area under the receiver operating characteristic curve Range 0, 1In Equations (1) and (2), N, y i ,y^ , andy¯represent the total number of observations, the actual value for the ith observation, the predicted value of y, and the average value of y, respectively. In Equations (3)–(9), the numbers of true negatives (TN), false negatives (FN), true positives (TP), and false positives (FP) are used as the inputs of metric.R2 is a statistical measure that represents the portion of the variance in the dependent variable that is predictable from the independent variable(s). The value ranges between 0 and 1, where 1 indicates that the model perfectly predicts the dependent variable. RMSE is a measure of the average magnitude of the error between predicted values and actual values. It is computed by obtaining the square root of the mean squared error. R2 and RMSE are both commonly used to evaluate the performance of regression models. They can be used to evaluate the performance of a model that predicts the metabolism and excretion outcomes. High R2 and low RMSE values indicate that the model can accurately predict the metabolism and excretion outcomes, while low R2 and high RMSE values suggest that the model cannot accurately predict the metabolism and excretion outcomes.SE is the ratio of true positive predictions among all actual positive cases. SP is the proportion of true negative predictions among all actual negative cases. MCC is a measure influenced by all values including TP, TN, FP, and FN; it is a measure of the balance between true positives and true negatives. ACC is the proportion of correct predictions among all predictions. It measures the overall performance of the model. The Jaccard score, which is an intersection of union with respect to the minority class, is another obvious method to cope with unbalanced data [13]. Precision is the proportion of true positive predictions among all positive predictions. It measures the ability of the model to avoid false positive predictions. Recall (also known as sensitivity) is the proportion of true positive predictions among all actual positive cases. It measures the ability of the model to identify all the positive cases. The F1 score is the harmonic means of precision and recall. It is a measure that considers both precision and recall and it can be a better metric than precision and recall alone. AUC is a metric of a model’s ability to differentiate between positive and negative classes. AUC ranges from 0 to 1, where a value of 1 indicates that the model can perfectly distinguish between positive and negative cases, while a value of 0.5 indicates that the model is not able to distinguish between positive and negative cases. In AI-based drug metabolism and excretion prediction, these metrics are commonly used as evaluation metrics for classification problems.There is no single metric that is superior in all cases. When evaluating and comparing the performance of different AI-based models for drug metabolism and excretion prediction, it is important to use appropriate evaluation metrics, use multiple evaluation metrics, make sure when comparing the model that all of the metrics are the same, consider the variability and uncertainty of the model’s performance, and consider the real-world implications of the model performance. 3. Drug Metabolism PredictionDrug metabolism refers to the biochemical processes by which the body modifies and eliminates drugs and other foreign substances. The body metabolizes drugs to either activate or inactivate them, and this can influence their therapeutic effects and potential for toxicity. The process of drug metabolism reactions can be classified into two main phases based on their chemical nature [3,6] (Figure 2). Phase I metabolism typically involves the oxidation, reduction, or hydrolysis of the drug, which can produce metabolites that are either inactive, active, or toxic. This process is primarily carried out by a family of enzymes called cytochrome P450 (CYP) enzymes. Phase II metabolism involves the conjugation of the modified drug with another molecule, such as glucuronic acid, sulfate, or amino acids. This process increases the water solubility of the drug and makes it easier to excrete. Phase II reactions are typically carried out by a variety of enzymes, including UDP-glucuronosyltransferases (UGTs), sulfotransferases, and glutathione S-transferases (GSTs). It is crucial to keep in mind that these reactions do not have to happen in order; they could even occur in reverse, in phase II, then in phase I, or as a single reaction [14]. The rate and extent of drug metabolism can vary greatly depending on the specific drug and individual factors such as genetics, sex, age, and disease status. Some drugs may be metabolized very quickly, while others may be metabolized very slowly, leading to the accumulation of potentially toxic levels of the drug in the body.The human CYP family has 57 isozymes [15]. CYPs are the primary enzymes involved in the metabolism of drugs, accounting for approximately 75% of overall metabolism, with about 95% of this activity being attributed to five isozymes, including 1A2, 2C9, 2C19, 2D6, and 3A4 [16]. The CYP-mediated metabolism of a novel chemical entity is of great importance during drug development because it has the potential to significantly impact the compound’s initial bioavailability, desired activity, and safety profile [17]. Quantitatively, UGTs accounted for 14% of the total metabolites collected, second only to the occurrence of CYP-catalyzed reactions [18]. Therefore, any study that leads to further insight into the mechanical aspects of metabolism will significantly support drug candidates’ rational design.The majority of drug-related metabolism takes place in the liver, as the enzymes that facilitate the reactions are concentrated there. Some drugs can be inhibitors or inducers of metabolic enzymes. If one drug is an inhibitor of the metabolism of another drug, when the two drugs are taken together in the body, the exposure of the other drug may be higher than expected, leading to potential safety problems. If one drug is an inducer of an enzyme that metabolizes another drug, when the two drugs are used concurrently, the effect of the other drug may be lower than expected, leading to adverse potential pharmacological effects in the body. This phenomenon is commonly known as a drug–drug interaction [19]. Metabolism can also create metabolites that are good for medicine and toxic metabolites [18]. Therefore, enzymatic metabolism studies are used to resolve metabolic stability, quantify, and identify main metabolites, identify metabolic pathways, and assess the possibility of drug–drug interactions throughout the preclinical stage and drug discovery [20].In silico, AI applications in the field of metabolic prediction fall into three major categories: (1) the sites of metabolism (SOMs) prediction, (2) metabolite structures prediction, and (3) metabolic pharmacokinetics prediction [10,21]. We provide recent developments of AI models in each category in the sections that follow. 3.1. SOMs and Metabolite Structure PredictionsThe prediction of SOMs is critical for a xenobiotic since it gives critical information for the derivation of potential metabolites [22]. Chemists can usually predict the structure of a metabolite by knowing the atom position in the molecule where the metabolizing reaction is most likely to occur [23]. In silico approaches for predicting the SOMs and metabolite structures in CYP-mediated processes are commonly used as a starting point for metabolic pathway research, which can also help with drug/lead optimization. Much software to predict SOMs for phases I and II has been developed, such as FAME [24], FAME 2 [25], FAME 3 [26], GLORY [23], GLORYx [27], BioTransformer [28], CypReact [13], CyProduct [29], and PreMetabo [30], summarized in Table 1.CypReact software uses machine learning (ML) to predict when a small chemical will react with any of the nine critical CYP isozymes. It employs a random forest (RF) model for each of the seven isozymes (1A2, 2A6, 2B6, 2C8, 2C19, 2E1, 3A4) and ensemble models (RF, support vector machine (SVM), logistic regression, and decision tree) for the remaining two isozymes (2C9, 2D6). Each model predicts substrate specificity based on a set of structural features and physicochemical properties of a molecule. Authors used 679 compounds from XenoSite [31] and manually gathered 1053 unreacted compounds to enhance the quality and predictability of the dataset, including known medicines, pesticides, dietary components, pollutants, endogenous metabolites, and a range of other substances. CypReact’s classifiers produce extremely high performance, with AUC scores between 83 and 92%. Additionally, CypReact is statistically superior to the baseline, according to a simple paired t test, with p values < [4.17E–6, 2.60E–4, 2.36E–5, 1.46E–4, 4.41E–5, 5.01E–6, 3.23E–5, 6.44E–6, 3.25E–6] for the nine CYPs. With p values < [4.90E–6, 5.25E–6, 1.54E–7], CypReact is statistically superior to SMARTCyp [32] for all three of the studied isoforms.Developed by the same author team, CyProduct is an in silico metabolism predictor to accurately predict the byproducts of human CYP metabolism. It consists of three tools: (1) CypReact predicts whether the query compound reacts with a specific CYP enzyme; (2) CypBoM Predictor predicts the reaction’s “bond site”; and (3) MetaboGen produces metabolic byproducts based on the bond-site prediction of CypBoM. It predicted the metabolic biotransformation products of the nine most essential human CYP enzymes: 1A2, 2A6, 2B6, 2C8, 2C9, 2C19, 2D6, 2E1, and 3A4. CypBoM makes use of a novel notion called “bond of metabolism” (BoM), which complements the classic “site of metabolism” by identifying the collection of chemical bonds that are changed or created during a metabolic reaction. A BoM dataset for 1845 CYP-mediated phase I reactions was created, and it was used to train the CypBoM predictor to anticipate the reactive bond position on substrate molecules. The cross-validated Jaccard score generated by CypBoM Predictor ranged between 0.380 and 0.452 for reactive bond prediction for the nine CYP enzymes. CypBoM Predictor’s Jaccard score is 0.13 better than that of FAME 2 and 0.12 better than that of FAME 3 in terms of SOMs on the 86 compounds. Moreover, CyProduct surpassed the other software tools, including ADMET Predictor, GLORY, and BioTransformer, in predicting metabolites by an average of 200% across variants of a testing dataset of 68 CYP substrates and 30 non-reactants concerning Jaccard scores. More specifically, for the BioTransformer dataset, the performance of CyProduct is about 30% better than that of BioTransformer and ADMET Predictor. Among the above software packages, CyProduct software is the most recently published and has a relatively detailed and clear performance comparison with others. 
Table 1.Public metabolism prediction tools. Table 1.Public metabolism prediction tools. NameMetabolism PredictionMethodsWebsite *Ref.CyProduct (CypReact, CypBoM, MetaboGen) Reactant, BoM for CYP, metabolite structureMLhttps://bitbucket.org/wishartlab/cyproduct/src/master/[29]GLORYxMetabolite structureMLhttps://nerdd.univie.ac.at/gloryx/[27]FAME 3Phase 1 and 2 SOMs for CYPMLhttps://nerdd.univie.ac.at/fame3/[26]BioTransformer 3.0Metabolic transformationrule-based/knowledge-baseb, ML http://biotransformer.ca/[28]PreMetaboPhase 1 and 2 SOMs for CYP, UGT, and SULTArrhenius equation and EaMEAD modelhttps://premetabo.bmdrc.kr/[30]SMARTCyp 3.0SOMs for CYPrule-based http://smartcyp.sund.ku.d/[33]HelixADMETCYP inhibitors and substratesGNN https://paddlehelix.baidu.com/app/drug/admet/train[34]Interpretable-ADMETCYP inhibitors and substratesGAT, GCNNhttp://cadd.pharmacy.nankai.edu.cn/interpretableadmet/[35]FP-ADMETCYP inhibitors and substratesRFhttps://gitlab.com/vishsoft/fpadmet[36]ADMETlab 2.0CYP inhibitors and substratesGCNNhttps://admetmesh.scbdd.com/[37]AdmetSAR 2.0CYP inhibitors and substratesRF, k-NN, SVMhttp://lmmd.ecust.edu.cn/admetsar2/[38]SwissADMECYP inhibitorsMLR, RNN, SVMhttp://www.swissadme.ch/[39]ICDrug ADMETCYP inhibitors and substratesRFwww.icdrug.com/ICDrug/ADMET[40]Virtual RatCYP inhibitorsRFhttps://virtualrat.cmdm.tw/[9]DL-CYPCYP inhibitorsDNNhttp://www.pkumdl.cn/deepcyp/home.php[41]CYPstrateCYP substratesRF, SVMhttps://nerdd.univie.ac.at/cypstrate/[42]CYPlebrityCYP inhibitorsRFhttps://nerdd.univie.ac.at/cyplebrity/[43]SuperCYPsPredCYP inhibitorsRFhttp://insilico-cyp.charite.de/SuperCYPsPred/[44] * Websites were accessed on 15 October 2022. Abbreviations: k-NN: k-nearest neighbor, MLR: multiple linear regression, RNN: recurrent neural network, DNN: deep neural network, GCNN: graph convolutional network, GAT: graph attention network.FAME, FAME 2, GLORY, FAME 3, and GLORYx are extensive series of legacy metabolic prediction software, respectively. The latest version, GLORYx, has extended the approach from GLORY, which combines SOMs prediction with a collection of reaction rules to predict phase I and II metabolism. Researchers used the SOMs probabilities prediction by the FAME 3 ML model on the SOMs dataset containing 1748 parent molecules from MetXBioDB [28] and the DrugBank database to achieve the predicted metabolites. FAME 3 uses extremely randomized tree classifiers and circular descriptors, including 15 basic 2D CDK descriptors and circular atom-type fingerprints. On a curated test dataset collecting phase I and phase II metabolites, GLORYx achieved an AUC of 0.79 and a recall of 77%. This performance was better than that of the GyGMa tool [45] on the same dataset but not better when regarding only phase II metabolite prediction. Furthermore, the authors note that it is difficult to give a firm definition of the area of application of GLORYx due to the scarcity of available high-quality data on small-molecule metabolism. 3.2. CYP Inhibitor and Substrate PredictionAs mentioned above, the CYP enzyme family plays a crucial role in drug metabolism. CYP inhibitors can affect the metabolism of drugs by reducing the activity of the CYP enzymes involved in their metabolism, leading to changes in the pharmacokinetics and pharmacodynamics of the drugs. On the other hand, the rate and extent of CYP substrate metabolism can be influenced by the presence of CYP inhibitors or other factors. Predicting the potential of a drug to be a CYP inhibitor or substrate is a complex process that involves many factors and variables, including the individual’s genetic makeup, age, and overall health. As such, predictions may not always be accurate and may need to be confirmed through further testing and analysis.Moreover, especially in recent years, there have been a lot of studies focusing on the prediction of specific metabolic CYP isoforms with remarkably good performance. Many prominent studies have focused on predicting the 5 major CYP inhibitors (1A2, 2C19, 2C9, 2D6, and 3A4), DeepCYP [41], SuperCYPsPred [44], CYPlebrity [43], iCYP-MFE [46], VirtualRat [9], and others [47,48]. Some studies focused on CYP substrate prediction, such as [33,47,49]. Some studies have focused on predicting only one CYP, such as CYP1B1 [50], CYP1A2 [51], CYP2C8 [52], CYP2C9 [53,54], and CYP3A4 [55,56]. Their performances are summarized in detail in Table 2. Table 2.Summary of AI methods to predict CYP subtypes from 2019 to 2022. Table 2.Summary of AI methods to predict CYP subtypes from 2019 to 2022. CYP SubtypesMethodsData SourcesDataset Size(Compounds)Best PerformanceRef.1A2, 2C19, 2D6, 2C9 and 3A4 inhibitorsRFPubChem, SuperCYP 18,313ACC = 0.97, AUC = 0.98[44]1A2, 2C9, 2C19, 2D6 and 3A4 inhibitorsRFChEMBL, PubChem, ADME 134,844AUC = 0.92, ACC = 0.83[43]1A2, 2D6, 2C9, 2C8, 2C19, and 3A4 inhibitorsRF[52,57]17,652ACC = 0.868, AUC = 0.741[36]CYPs 1A2, 2C9, 2C19, 2D6 and 3A4 inhibitorsRF, SVM, k-NN PubChem65,467AUC =0.93[46]1A2, 2D6, 2C9, 2C19, and 3A4 inhibitors2D6, 2C9, and 3A4 substrateRF, SVM, k-NN[57]77,4902018ACC = 0.855, AUC = 0.84[38]1A2, 2D6, 2C9, 2C19, and 3A4 inhibitorsDT[58,59]64,129ACC = 0.93, Recall = 0.924[9]1A1, 1A2, 2A6, 2B6, 2C8, 2C9, 2C19, 2D6, 2E1, and 3A4 substratesImproved Bayesian methodSuperCYP [60], PubChem, DrugBank, CYP450 Engineering Database [61,62], Meta-CYP7114AUC = 0.92, ACC = 0.90[49]1A2, 2C19, 2C9, 2D6, and 3A4 inhibitors1A2, 2C19, 2C9, 2D6, and 3A4 substratesMGAFChEMBL, PubChem, OCHEM, literature62,771 (inhibitors)3291 (substrates)ACC = 0.886, AUC = 0.948[37]1A2, 2C19, 2C9, 2D6, and 3A4 inhibitors2C9, 2D6, and 3A4 substratesGCNN, GATChEMBL, PubChem, DrugBank, literature63,921 (inhibitors)2053 (substrates)ACC = 0.85, AUC = 0.93[35]1A2, 2C19, 2C9, 2D6, and 3A4 inhibitors1A2, 2C19, 2C9, 2D6, and 3A4 substratesGNNPubChem, CypReact [13], SuperCYP [44]64,801 (inhibitors)9233 (substrates)AUC = 0.967[34]1A2, 2D6, 2C9, and 2C19 inhibitorsRF, GBDT, XGB, DNN, CNN[41] 53,179ACC = 0.974, AUC = 0.991[63]1A2, 2C9, 2C19, 2D6 and 3A4 inhibitorMT-DNN PubChem153,484AUC = 0.937, ACC = 0.895[48]2C8 inhibitorsRF, SVM, k-NN, LR, ANNPubChem and literature [64,65]514AUC = 0.90, ACC = 0.89[52]2C9 inhibitorsRF, SVMChEMBL 8141ACC = 0. 843, MCC = 0.695[54]2C9 inhibitorsBT, multilayer feedforward of resilient backpropagation networkPubChem >35,000AUC = 0.85[53]3A4 inhibitorsGCNN combined with the MT-DNNChEMBL and [66]3774R2 = 0.692[67]89,619R2 = 0.4143A4 inhibitorsSVM, XGB, and RFIn-house, public30,768ACC = 0.927, sensitivity = 0.788[56]In-house26,138ACC = 0.90, AUC = 0.9081B1 inhibitorsRF, SVM, ANNChEMBL, Pubchem, and [68,69,70]714MCC = 0.95[50]1A1 inhibitors658MCC = 0.961A2 inhibitorsCNNPubChem 21,721ACC = 0.722, AUC = 0.819[51] Abbreviations: ANN: artificial neural networks, CNN: convolutional neural networks, GNN: graph neural network, MT-: multitask-, DT: decision trees, GBDT: gradient boosting decision trees, LR: linear regression, XGB: extreme gradient boosting, BT: boost tree, MGAF: a multi-task graph attention framework.SuperCYPsPred [44] is a free, friendly, and publicly accessible web application that employs well-established ML techniques to predict five key CYP inhibitors, including 1A2, 2C19, 2C9, 2D6, and 3A4. The model was constructed using RF and various types of data sample methods on a dataset of 1170 pharmaceuticals with Morgan and MACCS circular fingerprints from their in-house SuperCYP and PubChem databases. SuperCYPsPred is extremely accurate, with an average cross-validation ACC of 93% and an average external validation ACC of 88.2%. SuperCYPsPred is among the most effective free tools for CYP prediction, making it a good preclinical drug discovery and development screening tool.In addition, some large recently developed ADMET (absorption, distribution, metabolism, excretion, and toxicity) prediction tools also integrate CYP prediction with high efficiency, such as HelixADMET [34], Interpretable-ADMET [35], FP-ADMET [36], ADMETLab 2.0 [37], and admetSAR 2.0 [38]. Many other prominent tools are summarized in Table 1. We also summarize in detail in Table 2 the AI models developed since 2019 that focus on predicting CYP subtypes. It is important to note that while these methods can provide useful information, they are not always predictive of the effects seen in vivo, and further in vivo studies are needed to fully understand the impact of CYP inhibition and substrate interactions on drug metabolism and efficacy. 3.3. UGTs PredictionGlucuronosyltransferases are responsible for the glucuronidation process, a primary and most important part of phase II metabolism [71]. The UGT enzyme [15] catalyzes the addition of a glucuronic acid moiety to xenobiotics, which is the primary method through which the human body eliminates the most frequently prescribed medications. It is also the primary route of chemical elimination for the majority of medications, dietary agents, poisons, and endogenous compounds from the diet, environmental sources, and pharmaceutical industries. However, compared with phase I reactions, phase II metabolism is much less noticeable, although it has an important impact on modulating pharmacological effects [72]. Predicting the potential of a drug to be metabolized by UGTs is an important aspect of drug development, as it can help to determine the risk of drug–drug interactions and potential adverse effects. Table 3 summarizes several studies predicting UGTs in recent years.Mazzolari et al. built two models using molecular descriptors and RF algorithms to predict UGT-mediated metabolism [73]. The first model predicts whether a molecule is prone to conversion to glucuronide using 2192 molecules from the MetaQSAR database [74], achieving an AUC of 0.94 and an MCC of 0.76 in the internal evaluation and an AUC of 0.90 and an MCC of 0.70 in external evaluation using 120 additional xenobiotics. The second model differentiates between the two major forms of glucuronidation by determining whether conjugation takes place on an oxygen or nitrogen atom (O- or N-glucuronidation) using 661 O-glucuronidation and 114 N-glucuronidation substrates, with a recall value of 0.78. This result emphasizes the need to utilize well-curated datasets when developing new methodologies for predicting phase II metabolism and demonstrates a practical application of the MetaQSAR database.The SOM prediction model was developed by Cai et al. [75] for four subtypes of UGT-mediated reactions, including AlOH, ArOH, COOH, and nitrogen, using DT, RF, and AdaBoost methods. They used 400 drugs metabolized by UGT from the Handbook of Metabolic Pathways of Xenobiotics [76] and two external test sets from previous studies [77,78,79]. Differently sized atom environment fingerprints were used to describe the SOMs. The best performance of their optimal models yielded an ACC of 86.7% on the test set and 79.8% on the external test sets. However, the use of small and undiversified datasets is one of their limitations, which hinders the full exploitation of the possibilities of ML methods and the performance is not appreciated.PreMetabo [30] is an available web tool to predict phases I and II drug metabolism using knowledge-based prediction models. For phase I drug metabolism prediction, Hwang et al. used the EaMEAD model based on the Arrhenius equation for four CYP enzymes (1A2, 2C9, 2D6, 3A4) to predict the SOMs on the drug molecule. In phase II, they developed a consensus classification model using SVM to predict the UGT and SULT substrate. PreMetabo used 200 substrates of each CYP from Fujitsu ADME database for evaluation and comparison with the BioTransformer tool [28]. The predictability of the primary metabolite in the top-3 was determined to be from 72.5 to 84.5% for four CYPs in the SOMs prediction model. The PreMetabo recall value of all CYPs was higher than the BioTransformer recall value on the same dataset and it was judged to be more practical than BioTransformer. Besides, in phase II, they used UGT and SULT substrates from MDL metabolism and Fujitsu ADME databases with 1024 ECFP4 fingerprints and 881 PubChem fingerprints. The highest accuracy of their models was determined to be 93.9 and 80.7%, respectively, for internal validation. Moreover, PreMetabo achieved a UGT substrate prediction ACC of 81% on the external test dataset containing 11 FDA-approved drugs. Table 3.Summary of AI methods to predict UGT property from 2019 to 2022. Table 3.Summary of AI methods to predict UGT property from 2019 to 2022. MethodsData SourcesDataset Size(Compounds)PerformanceRef.RFMetaQSAR7962MCC = 0.76, AUC = 0.94[73]DT, RF, AdaBoostHandbook of Metabolic Pathways of Xenobiotics [76], KEGG [77,78,79]586ACC = 0.867, AUC = 0.928[75]LR, SVMFujitsu ADME database and MDL metabolism database200ACC = 0.81 [30]4. Drug Excretion PredictionDrug excretion is the process through which medications are excreted from the body, as metabolites or as original drugs [4]. Excretion is a complicated process involving many elimination routes. The kidneys are in charge of excreting the majority of chemicals that are water-soluble. Additionally, the biliary system can excrete medications that are not absorbed by the stomach tract. The number of drugs eliminated by the intestines, saliva, sweat, breast milk, and lungs is usually insignificant. On the other hand, certain volatile anesthetics are capable of being exhaled via the lungs. Additionally, even minute amounts of the drug in a nursing woman’s breast milk can have an effect on her nursing infant. During development, drug excretion properties contribute to the validation of toxicity studies, aid in assessing safety before the first dose in humans, provide dosimetry data in humans for the clinical, and indicate the possibility of drug–drug interactions. The main pharmacokinetic parameters for drug excretion include clearance and half-life (t1/2). In the next section, we present recent advances in predictive research clearance and t1/2 properties using AI. 4.1. Clearance PredictionThe volume of plasma cleared of a drug over a given time period is referred to as drug clearance [80]. As a result, the unit of measurement for drug clearance is volume/time. Another equation can be used to compute the discharge. Clearance is computed by dividing the rate of elimination of a drug from plasma (mg/min) by its concentration in plasma (mg/mL). The entire ability of the body to eliminate the medication from plasma is comprised of renal clearance, hepatic clearance, and clearance from all other tissues. Clearance may be affected by body weight and surface area, cardiac output, renal function, liver function, plasma protein binding, concomitant medications, and changed expression presence of drug-metabolizing enzymes [81]. Clearance is a critical pharmacokinetic parameter to consider in both drug discovery and clinical practice because clearance is a factor of all other relevant pharmacokinetic parameters, including half-life, oral bioavailability, and effective dose [82]. Many recent studies predicting clearance property in silico are summarized in Table 4.Recently, software has been developed to predict ADMET properties named FP-ADMET, which integrates clearance prediction including human renal clearance, intrinsic clearance, metabolic intrinsic clearance, and human liver microsomal clearance [36]. Researchers used a fingerprint-based RF algorithm for the four-clearance prediction models. The data used to evaluate the four models were 636 compounds, 244 compounds, 5278 compounds, and 5348 compounds, respectively, which were collected from many previous studies [66,83,84,85]. The human renal clearance prediction model of FP-ADMET gave better predictive results than the study of Chen et al. [85] on the same dataset and algorithm with R2 of 0.27 and RMSE of 0.53 (compared to R2 = 0.2 and RMSE = 1.8). However, the intrinsic clearance prediction model of FP-ADMET showed no better predictive results than the study of Hsiao et al. [84] on the same dataset with R2 of 0.29 (compared to R2 = 0.96). The accuracy of the metabolic intrinsic clearance model by FP-ADMET was 74%, higher than the accuracy of the RF model used by Esaki et al. [83] on the same dataset (72.3%), but the radial SVM model used by Esaki et al. had a higher ACC of 77.1%. The performance of the human liver microsomal clearance model by FP-ADMET was no better than that of the MT-DNN model used by Wenzel et al. [66] on the same dataset, with an R2 of 0.56 (compared to R2 = 0.624). The RF model developed by Wang et al. for the clearance prediction also achieved higher performance than other models such as SVM, GBM, and XGB on 1352 compounds, with an R2 of 0.875 and an RMSE of 0.103 [86]. Furthermore, Kosugi and Hosea once again proved that the total plasma clearance prediction model using the RF algorithm is more efficient than many other algorithms, such as radial basis function fitting (RBF), partial least squares (PLS), random forest regression (RFR), Gaussian process models (GP) with two-dimensional search for parameters (GP2DS), fixed hyperparameters (GPFixed), hyperparameters obtained by forward variable selection (GPFVS), rescaled procedure (GPRFVS), and by conjugate gradient optimization (GPOPT) on the same dataset of 1114 compounds with an RMSE of 0.4 using five-fold cross-validation [87]. The best performing human renal clearance prediction model developed by Watanabe et al. is also a model using the RF algorithm when compared with other algorithms such as SVM, PLS, and ANN on 401 compounds, with an R2 of 0.92 and an RMSE of 0.12 [88]. With many years of experience in ADMET prediction and a large internal dataset of 73,620 compounds, AstraZeneca built a clearance prediction model using the SVM algorithm, with good results with an RMSE of 0.377 [89].In addition to ML algorithms, recently deep learning (DL) algorithms have also been exploited and built predictive models of clearance with remarkable efficiency. Mamada et al. successfully combined conventional ML using molecular descriptors with DeepSnap-DL to build a new clearance prediction model [90]. They used rat clearance data containing 1545 in-house compounds to evaluate the prediction performance. With an AUC and an ACC of 94.3 and 87.4%, respectively, their ensemble model did better than conventional ML (AUC = 88.3% and ACC = 82.5%) or DeepSnap-DL (AUC = 90.5% and ACC = 83.2%). Sohlenius-Sternbeck et al. developed an intrinsic clearance prediction model using an ANN algorithm and 4794 compounds from Medivir in-house dataset [91]. This model was a significant improvement over ADMET PredictorTM from Simulations Plus, with R2 of 0.717 (compared to R2 = 0.53). Using the same dataset of 5384 compounds, the combined model of GCNN and MT-DNN model of Liu et al. [67] and the MT-DNN model of Wenzel et al. [66] achieved approximately equal accuracy in clearance prediction (R2 = 0.62). Recently, DL technical was also exploited in ADMETLab 2.0 using MGAF to predict clearance on 831 compounds, achieved an R2 of 0.629 [14]. Table 4.Summary of AI methods to predict clearance property from 2019 to 2022. Table 4.Summary of AI methods to predict clearance property from 2019 to 2022. MethodsData SourcesDataset Size(Compounds)PerformanceRef.RFHuman renal clearance [85]636R2 = 0.27, RMSE = 0.53[36]Intrinsic clearance [84]244R2 = 0.29, RMSE = 1.02Metabolic intrinsic clearance [83]5278ACC = 0.74, AUC = 0.84Human liver microsomal clearance [66]5348R2 = 0.56, RMSE = 1.05RF, SVM, GBM, XGB[92]1352R2 = 0.875, RMSE = 0.103[86]RFR, RBF, PLS, GP2DS, GPFixed, GPFVS, GPRFVS, GPOPTTakeda Pharmaceutical Company (Fujisawa, Japan)1114R2 = 0.61, RMSE = 0.31[87]SVMAstraZeneca in-house data73,620R2 = 0.356, RMSE = 0.377[89]RF, NB, SVM, CT, k-NN, MLR, ANNFDA drugs and [93,94,95,96]636R2 = 0.94, RMSE = 0.11[85]RF, AdaBoost, Radial SVM, Linear SVMChEMBL v.23, KEGG DRUG [97]56,065 ACC = 0.77, Kappa = 0.588[83]RF, SVM, PLS, ANNChEMBL and Varma et al. [98]401R2= 0.92, RMSE = 0.12[88]Combination conventional ML and DeepSnap-DLin-house 1545AUC = 0.943, ACC = 0.874[90]ANNMedivir in-house 4794R2 = 0.717, RMSE = 0.327[91]GCNNChEMBL, PubChem, OCHEM, literature831R2 = 0.692[37]a molecular GCNN combined with the MT-DNN[66]5348R2 = 0.62[67]Amgen’s internal datasets86,470R2 = 0.445MT-DNNChEMBL v.235384R2 = 0.624[66]MT-CNNAstraZeneca 139,907R2 = 0.59, RMSE = 0.35[99] Abbreviations: NB: Naïve Bayes, CT: classification tree, SVR: support vector regression. 4.2. Half-Life PredictionThe excretion half-life of the drug is the length of time needed for the amount of the active component in the drug to decrease by half of its starting dose in the body [100]. This is dependent on how the substance is metabolized and eliminated by the body. It can last anywhere from a few hours to several days or even weeks. Understanding the concept of half-life makes it possible to calculate the steady-state concentrations and excretion rates for any given drug. A more frequent dosage may be necessary to maintain the proper level of exposure and prevent unnecessary peak concentrations if a drug’s half-life is too short [101]. As a result, it could be more challenging to achieve the best efficacy, safety, and patient compliance. A drug’s extremely lengthy half-life may increase the amount of time needed for subsequent accumulation and elimination. This can complicate the management of adverse events and the design of efficient clinical trials. An accurate estimate of the time needed for medicine or substance to be excreted from the body is difficult to come by. Some of the models developed since 2019 are summarized in Table 5.Interpretable-ADMET, a new ADMET predictor, uses GCNN and GAT algorithms to predict 59 ADMET properties, including half-life [35]. The GAT model gave a slightly better half-life prediction result than the GCNN model, with an ACC of 77.6 and 77.3%, respectively, on 665 compounds. In ADMETLab 2.0, a multi-task graph attention framework was used to build the ADMET prediction models, including half-life prediction model [37]. The half-life prediction model was evaluated on 1219 compounds and had a predictive ACC of 74% and an AUC of 82%. Furthermore, the software that has been developed for predicting ADMET properties named FP-ADMET also has integrated half-life prediction [36]. Researchers used a fingerprint-based RF algorithm and 2127 compounds from MetStabOn [102] to predict half-life. When using the same dataset and RF algorithm, FP-ADMET predicts half-life more accurately than MetStabOn, with an ACC of 76 and 72.6%, respectively. In another study, Wang et al. built a predictive model of the half-life and three other properties using ML methods, including RF, gradient boosting machines (GBM), SVM, and XGB [86]. They used a dataset of 1352 compounds from Lombardo et al. [92] and 162 critical variables, including 2D molecular, 3D molecular, and fingerprint descriptors. Assessed by 10-fold cross-validation, the RF model produced more accurate prediction than other models, with an R2 of 0.832 and an RMSE of 0.154. The actual half-life of the same drug can differ considerably between individuals due to a variety of patient- and drug-specific characteristics. Therefore, in silico studies predicting drug half-life are also very limited. There is hardly a single recent half-life predictive study. Most half-life prediction models are mainly integrated into large ADMET prediction programs or researched with many other properties.The excretion of a drug is a complicated process involving many elimination pathways, including biliary excretion, renal excretion, and others, each of which includes many different processes [22]. Till now, in silico excretion predictors have been difficult to develop due to the complex drug excretion processes. Table 5.Summary of AI methods to predict half-life property from 2019 to 2022. Table 5.Summary of AI methods to predict half-life property from 2019 to 2022. MethodsData SourcesDataset Size(Compounds)PerformanceRef.RF[102]2127ACC = 0.76, AUC = 0.88[36]SVM, RF, GBM, XGB[92]1352R2 = 0.832, RMSE = 0.154[86]MGAFChEMBL, PubChem, OCHEM, literature1219AUC = 0.822, ACC = 0.744 [37]GCNN, GATChEMBL, PubChem, DrugBank, literature 665ACC = 0.773, AUC = 0.766[35] Although researchers have proposed many AI-based models to predict drug metabolism and elimination, evaluating and comparing them on an objective basis can be quite challenging. A lack of consensus datasets and evaluation metrics can be a major limitation in comparing predictive models in the field of biology and biomedicine. In Table 2, Table 3, Table 4 and Table 5, we summarize recently developed AI-based methods for drug metabolism and elimination prediction. We only provide comparative information when the authors used the same dataset and metrics to evaluate their models in the content. 5. Data Sources for Research CommunityThe selection of an appropriate database is a critical step in the development of accurate and reliable AI-based predictive models for metabolism and excretion. Careful consideration should be given to the quality, completeness, and relevance of the data in order to ensure the best possible results. Some commonly used databases to predict drug metabolism and excretion are briefly described as follows:HMDB 5.0 (https://hmdb.ca/ accessed on 22 January 2023): An extensive database of small molecule metabolites discovered in the human body, including information on their chemical and physical properties, metabolic pathways, and clinical biomarkers. Information on more than 220,000 metabolites and 8500 protein sequences can be found in HMDB. [103].METLIN (https://metlin.scripps.edu/ accessed on 22 January 2023): a metabolite database that contains information on more than 960,000 compounds [104]. It includes information on the chemical structure, molecular formula, and biological activities of metabolites. METLIN offers MS/MS data on various collision energy values in both positive and negative ionization modes. Additionally, it makes use of the elemental makeup, precise mass measurements, and the known structure of the metabolite to estimate the fragmented structure. The metabolomics-specific mobile interface METLIN Mobile allows you to see metabolite information from any cellular device.MetaCyc (https://metacyc.org/ accessed on 23 January 2023): A curated database of metabolic pathways and enzymes for a range of organisms. It includes information on 3085 pathways, 18,785 metabolites, and 18,391 reactions involved in metabolite biotransformation and can be used to construct metabolic models for specific organisms.MetaQSAR: A database for metabolites including information on the relationship between the chemical structure of a metabolite, its biological activity, the physicochemical properties of chemicals, as well as their predicted metabolic pathways and associated enzymes. It is a plug-in embedded in the VEGA ZZ programs (http://www.vegazz.net/ accessed on 23 January 2023) and contains 1890 substrates [74].MetXBioDB (https://bitbucket.org/djoumbou/biotransformerjar/src/master/ accessed on 23 January 2023): A database of metabolic pathways and enzymes for a range of organisms, including bacteria, archaea, and eukaryotes. MetXBioDB contains data on more than 2000 biotransformation including information on the structure and function of enzymes, as well as the reactions and pathways involved in metabolite biotransformation [28].Metabolights (https://www.ebi.ac.uk/metabolights/ accessed on 24 January 2023): A database of metabolomic data, which includes information on metabolites, metabolic pathways, and metabolic networks of more than 27,500 compounds. Metabolights also includes tools for data analysis and visualization, as well as resources for sharing and reusing metabolomic data [105].KEGG Pathway (https://www.genome.jp/kegg/pathway.html accessed on 24 January 2023): A database of metabolic pathways, including maps and diagrams of metabolic networks, as well as information on enzymes and metabolites. It includes information on more than 17,000 metabolic pathways and over 22,000 enzymes [106].HumanCyc (https://humancyc.org/ accessed on 24 January 2023): A curated database of metabolic pathways, enzymes for human metabolism, and the human genome. HumanCyc includes information on the reactions and pathways involved in metabolite biotransformation, as well as the enzymes and genes involved in these processes. Information on 28,783 genes, their products, and the metabolic processes and pathways they catalyze is contained in the pathway/genome database that was created as a consequence [107].BiGG (http://bigg.ucsd.edu/ accessed on 24 January 2023): In order to simulate systems biology and predict metabolic flux balance, the BiGG database reconstructs human metabolism metabolically. The 1496 ORFs, 2004 protein complexes, 2766 metabolites, and 3311 metabolic and transport processes are all included in this thorough literature-based genome-scale metabolic reconstruction. It was put together from building 35 of the human genome [108].DrugBank (http://www.drugbank.ca/ accessed on 24 January 2023): A comprehensive database of drug and drug target information including information on drug metabolism and pharmacokinetics, as well as the enzymes involved in drug biotransformation. It contains information on more than 500,000 drugs and their associated targets, pathways, and metabolic pathways [109].ChEMBL (www.ebi.ac.uk/chembl/ accessed on 24 January 2023): A database of bioactive molecules, including drugs and drug candidates, with information on their activities, targets, and metabolic pathways. It contains data on more than 2.3 million compounds and their associated activities and targets [110].ChemSpider (http://www.chemspider.com/ accessed on 24 January 2023): A chemical structure database that includes information on more than 115 million compounds including information on chemical structures, properties, and associated metadata, such as chemical identifiers and references [111].PubChem (https://pubchem.ncbi.nlm.nih.gov/ accessed on 25 January 2023): A public database of chemical structures and their associated biological activities including information on more than 114 million compounds, as well as tools for data analysis and visualization [112].ZINC20 (https://zinc20.docking.org/ accessed on 25 January 2023): A database of commercially available compounds for drug discovery including information on more than 750 million purchasable compounds, as well as tools for searching and filtering compounds based on various criteria, such as molecular weight, bioavailability, and toxicity [113].OCHEM (https://ochem.eu/home/show.do accessed on 25 January 2023): A platform for the development and validation of predictive models for chemical and biological data. OCHEM includes tools for data preprocessing, feature selection, and model training, as well as a library of pre-trained models. OCHEM contains more than 3.7 million records for 689 properties [114].Therapeutics Data Commons (TDC) (https://tdcommons.ai/ accessed on 25 January 2023): A database of clinical trial data for FDA-approved drugs including information on drug pharmacokinetics, pharmacodynamics, and adverse events, as well as data on drug metabolism and excretion. TDC contains data on more than 4.2 million compounds, 34,000 genes, and approximately 2 million reactions [115].openFDA (https://open.fda.gov/ accessed on 25 January 2023): A database of FDA-approved drugs, including information on drug labeling, adverse events, and clinical trial data. OpenFDA includes tools for data analysis and visualization, as well as an API for accessing FDA data [116].These databases and datasets are publicly available, so they can be valuable resources for researchers working on metabolism and excretion prediction, pharmacokinetic property prediction, drug discovery, and related areas. 6. Challenges in Drug Metabolism and Excretion Prediction Based on AIAI-based drug metabolism and excretion prediction presents the possibility of revolutionizing drug R&D, but there are still several issues that must be resolved to raise the accuracy and reliability of predictive models.Metabolism and excretion are complicated biological processes involving multiple enzymes, transporters, biochemical pathways, multiple organs, and other molecular components. The interaction of these various components can be difficult to accurately model, resulting in inaccurate predictions. Predicting a drug’s metabolism necessitates an understanding of the relevant pathways and enzymes. Human metabolism and excretion are very diverse and are influenced by a variety of factors, including age, gender, heredity, illness status, interacting drugs, dose, and the route of administration [117]. Accurately predicting properties for all individuals is difficult. Genetic variations in these enzymes can lead to differences in drug metabolism and elimination between individuals, which can make it difficult to predict how a drug will be metabolized and excreted in different populations. Many drugs have complex metabolic processes that involve multiple enzymes and pathways. These pathways are frequently interrelated, and predicting the activity of a single enzyme or pathway may not be sufficient to predict overall metabolism and excretion of a drug. Because of the existence of metabolic intermediates that would allow for intramolecular rearrangement, it is uncertain that the basic mechanism and regulations of drug metabolism can be characterized only based on the drug structures [15]. Nonlinearities in the input rate (for example, formation) and the output rate (for example, elimination) can all have an impact on the distribution of metabolites [118]. The mechanisms underlying drug metabolism and excretion are not fully understood, particularly for some drug classes, which can hinder the development of accurate prediction models.The lack of high-quality data is a significant challenge in developing accurate metabolism and excretion prediction models as well as AI-based drug discovery models in general [119]. Poor or incomplete data as well as erroneous data collection and analysis methods will lead to inaccurate predictions as the quality of the output is controlled by the quality of the input. AI algorithms require large amounts of comprehensive and high-quality experimental data to train and validate prediction models. However, generating comprehensive experimental data that accurately reflects these complex processes can be difficult. Experiments to measure drug metabolism and excretion can be time-consuming and costly, requiring specialized equipment and expertise [5]. As a result, there may be insufficient resources to generate the required data for all drugs of interest. In experimental studies, a lack of standardization and quality control measures can lead to variability and errors in the data, affecting the accuracy of AI models. The chemical diversity of compounds can make it challenging to develop universal prediction models that can accurately predict metabolism and excretion properties for a broad range of compounds. Data on drug metabolism and excretion can be collected from a variety of sources, including in vitro and in vivo experiments, clinical trials, and literature sources. These data sources may use different experimental methods, formats, and standards, and may have varying levels of quality, which can make it challenging to combine and analyze the data. Additionally, if the training data is limited or unrepresentative of the broader population, overfitting can occur, resulting in poor generalization of new data, and posing a challenge to the development of AI models for predicting drug metabolism and excretion.Transparency and interpretability of AI models to predict drug metabolism and excretion are important factors in ensuring the safety and efficacy of these models and their usefulness in clinical applications. However, achieving transparency and interpretability in AI models for drug metabolism and excretion prediction can be challenging, particularly given the complexity of the underlying biological processes. Interpretable AI applications should have desirable features such as transparency, justification, informativeness, and uncertainty estimation [120]. However, many AI models, particularly DL models, are considered “black box” models and are often difficult for human experts to interpret [120,121] because they are highly complex, with multiple layers and nonlinear interactions between different components. Model interpretability is dependent on the chemical representation and AI strategy of choice [122]. Full comprehension of DL models in the context of drug R&D may be challenging to attain, but the supplied predictions might still be helpful to the researcher. 7. Conclusions and Future DirectionThe development of AI models for drug metabolism and excretion prediction holds great promise for improving drug R&D. Researchers are working hard to explore new ways to create and integrate experimental data, such as relying on metadata to improve data quality [123]. In addition, efforts are underway to improve the quality of experimental data through standardization of experimental protocols, and the use of quality control measures and rigorous validation procedures. The integration of multi-omics data [124], such as genomics, transcription, proteomics, and metabolism, and integration with pharmacokinetic and toxicological modeling will allow for more comprehensive predictions of drug metabolism and excretion. This will allow for a better understanding of the molecular mechanisms underlying ADMET processes and the development of more accurate predictive models. The creation of common data standards and protocols, as well as networks and platforms for sharing data, makes it easier for different institutions and organizations to work together and share data. Collaboration and data sharing among researchers, pharmaceutical companies, and regulatory agencies can aid in the improvement of data quality and availability for AI model development. Model sharing can also help with the validation and testing of AI models in various contexts. AI models for drug metabolism and excretion prediction may be integrated with electronic health records to allow for more personalized medicine by considering individual patient characteristics such as genetic information, age, sex, and medical history. In the context of complex big data, DL methods are likely to prevail soon as they are easier to adapt to a wider range of chemical entities and modeling tasks and enable more efficient data mining. In addition to using existing explanatory AI methods such as feature attribution, instance-based, graph-convolution-based, and self-explanatory methods [123], efforts are being made to develop new methods to ensure transparency, safety, efficacy, and reliability in clinical settings, and maintain public trust in AI technology. In-depth knowledge of drug metabolism and excretion and AI techniques is very important to give a reasonable and useful explanation.Overall, we provided a comprehensive overview of recent AI-based drug metabolism and excretion prediction research, along with key challenges and future directions. A collaboration effort between AI experts, data scientists, chemists, biologists, and other related field experts and the integration of emerging technologies will be essential to realizing the full potential of this field.Author ContributionsConceptualization, T.T.V.T.; methodology, T.T.V.T.; validation, T.T.V.T., H.T. and K.T.C.; resources, K.T.C.; writing—original draft, T.T.V.T.; writing—review and editing, H.T. and K.T.C.; visualization, T.T.V.T. and H.T.; supervision, K.T.C.; project administration, H.T. and K.T.C.; funding acquisition, K.T.C. All authors have read and agreed to the published version of the manuscript.FundingThis work was supported in part by a National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIT) (No. 2020R1A2C2005612) and (No. 2022R1G1A1004613).Institutional Review Board StatementNot applicable.Informed Consent StatementNot applicable.Data Availability StatementNot applicable.Conflicts of InterestThe authors declare no conflict of interest.ReferencesTalevi, A.; Bellera, C.L. Drug excretion. In ADME Processes in Pharmaceutical Sciences; Talevi, A., Quiroga, P., Eds.; Springer: Cham, Switzerland, 2018. [Google Scholar]Ha, C.-E.; Bhagavan, N.V. Chapter 35—Drug metabolism. In Essentials of Medical Biochemistry, 3rd ed.; Ha, C.-E., Bhagavan, N.V., Eds.; Academic Press: Cambridge, MA, USA, 2023; pp. 795–805. [Google Scholar]Susa, S.T.; Preuss, C.V. Drug Metabolism; Statpearls: Treasure Island, FL, USA, 2022. [Google Scholar]Barreto, E.F.; Larson, T.R.; Koubek, E.J. Drug Excretion; Elsevier: Amsterdam, The Netherlands, 2021. [Google Scholar]Kirchmair, J.; Göller, A.H.; Lang, D.; Kunze, J.; Testa, B.; Wilson, I.D.; Glen, R.C.; Schneider, G. Predicting drug metabolism: Experiment and/or computation? Nat. Rev. Drug Discov. 2015, 14, 387–404. [Google Scholar] [CrossRef] [PubMed]Gad, S.C.; Spainhour, C.B. Nonclinical Drug Administration: Formulations, Routes and Regimens for Solving Drug Delivery Problems in Animal Model Systems; CRC Press: Boca Raton, FL, USA, 2017. [Google Scholar]Mackenzie, P.I.; Somogyi, A.A.; Miners, J.O. Advances in drug metabolism and pharmacogenetics research in Australia. Pharmacol. Res. 2017, 116, 7–19. [Google Scholar] [CrossRef]Lai, Y.; Chu, X.; Di, L.; Gao, W.; Guo, Y.; Liu, X.; Lu, C.; Mao, J.; Shen, H.; Tang, H. Recent advances in the translation of drug metabolism and pharmacokinetics science for drug discovery and development. Acta Pharm. Sin. B 2022. [Google Scholar] [CrossRef] [PubMed]Hsiao, Y.; Su, B.H.; Tseng, Y.J. Current development of integrated web servers for preclinical safety and pharmacokinetics assessments in drug development. Brief Bioinform. 2021, 22. [Google Scholar] [CrossRef] [PubMed]Kazmi, S.R.; Jun, R.; Yu, M.-S.; Jung, C.; Na, D. In silico approaches and tools for the prediction of drug metabolism and fate: A review. Comput. Biol. Med. 2019, 106, 54–64. [Google Scholar] [CrossRef] [PubMed]Smith, G.F. Artificial intelligence in drug safety and metabolism. Artif. Intell. Drug Des. 2022, 483–501. [Google Scholar] [CrossRef]Rifaioglu, A.S.; Atas, H.; Martin, M.J.; Cetin-Atalay, R.; Atalay, V.; Doğan, T. Recent applications of deep learning and machine intelligence on in silico drug discovery: Methods, tools and databases. Brief. Bioinform. 2019, 20, 1878–1912. [Google Scholar] [CrossRef]Tian, S.; Djoumbou-Feunang, Y.; Greiner, R.; Wishart, D.S. Cypreact: A Software Tool for in Silico Reactant Prediction for Human Cytochrome P450 Enzymes. J. Chem. Inf. Model. 2018, 58, 1282–1291. [Google Scholar] [CrossRef]Phang-Lyn, S.; Llerena, V.A. Biochemistry, biotransformation. In Statpearls [Internet]; StatPearls Publishing: Treasure Island, FL, USA, 2022. [Google Scholar]Li, Y.; Meng, Q.; Yang, M.; Liu, D.; Hou, X.; Tang, L.; Wang, X.; Lyu, Y.; Chen, X.; Liu, K. Current trends in drug metabolism and pharmacokinetics. Acta Pharm. Sin. B 2019, 9, 1113–1144. [Google Scholar] [CrossRef]Guengerich, F.P. Cytochrome P450 and chemical toxicology. Chem. Res. Toxicol. 2008, 21, 70–83. [Google Scholar] [CrossRef]Zheng, M.; Luo, X.; Shen, Q.; Wang, Y.; Du, Y.; Zhu, W.; Jiang, H. Site of metabolism prediction for six biotransformations mediated by cytochromes P450. Bioinformatics 2009, 25, 1251–1258. [Google Scholar] [CrossRef] [PubMed]Testa, B.; Pedretti, A.; Vistoli, G. Reactions and enzymes in the metabolism of drugs and other xenobiotics. Drug Discov. Today 2012, 17, 549–560. [Google Scholar] [CrossRef] [PubMed]Zhang, Z.; Tang, W. Drug metabolism in drug discovery and development. Acta Pharm. Sin. B 2018, 8, 721–732. [Google Scholar] [CrossRef] [PubMed]Mann, B.; Melton, R.; Thompson, D. Drug metabolism in drug discovery and preclinical development. In Drug Metabolism; IntechOpen: London, UK, 2021. [Google Scholar]Litsa, E.E.; Das, P.; Kavraki, L.E. Machine learning models in the prediction of drug metabolism: Challenges and future perspectives. Expert Opin. Drug Met. 2021, 17, 1245–1247. [Google Scholar] [CrossRef]Yang, X.; Wang, Y.; Byrne, R.; Schneider, G.; Yang, S. Concepts of Artificial Intelligence for Computer-Assisted Drug Discovery. Chem. Rev. 2019, 119, 10520–10594. [Google Scholar] [CrossRef] [PubMed]de Bruyn Kops, C.; Stork, C.; Sicho, M.; Kochev, N.; Svozil, D.; Jeliazkova, N.; Kirchmair, J. GLORY: Generator of the Structures of Likely Cytochrome P450 Metabolites Based on Predicted Sites of Metabolism. Front Chem. 2019, 7, 402. [Google Scholar] [CrossRef] [PubMed]Kirchmair, J.; Williamson, M.J.; Afzal, A.M.; Tyzack, J.D.; Choy, A.P.; Howlett, A.; Rydberg, P.; Glen, R.C. FAst MEtabolizer (FAME): A rapid and accurate predictor of sites of metabolism in multiple species by endogenous enzymes. J. Chem. Inf. Model. 2013, 53, 2896–2907. [Google Scholar] [CrossRef]Sicho, M.; de Bruyn Kops, C.; Stork, C.; Svozil, D.; Kirchmair, J. FAME 2: Simple and effective machine learning model of cytochrome P450 regioselectivity. J. Chem. Inf. Model. 2017, 57, 1832–1846. [Google Scholar] [CrossRef]Sicho, M.; Stork, C.; Mazzolari, A.; Kops, C.D.; Pedretti, A.; Testa, B.; Vistoli, G.; Svozil, D.; Kirchmair, J. FAME 3: Predicting the Sites of Metabolism in Synthetic Compounds and Natural Products for Phase 1 and Phase 2 Metabolic Enzymes. J. Chem. Inf. Model. 2019, 59, 3400–3412. [Google Scholar] [CrossRef]de Bruyn Kops, C.; Sicho, M.; Mazzolari, A.; Kirchmair, J. GLORYx: Prediction of the Metabolites Resulting from Phase 1 and Phase 2 Biotransformations of Xenobiotics. Chem. Res. Toxicol. 2021, 34, 286–299. [Google Scholar] [CrossRef]Djoumbou-Feunang, Y.; Fiamoncini, J.; Gil-de-la-Fuente, A.; Greiner, R.; Manach, C.; Wishart, D.S. Biotransformer: A comprehensive computational tool for small molecule metabolism prediction and metabolite identification. J. Cheminform. 2019, 11, 2. [Google Scholar] [CrossRef] [PubMed]Tian, S.; Cao, X.; Greiner, R.; Li, C.; Guo, A.; Wishart, D.S. Cyproduct: A Software Tool for Accurately Predicting the Byproducts of Human Cytochrome P450 Metabolism. J. Chem. Inf. Model. 2021, 61, 3128–3140. [Google Scholar] [CrossRef] [PubMed]Hwang, S.; Shin, H.K.; Shin, S.E.; Seo, M.; Jeon, H.N.; Yim, D.E.; Kim, D.H.; No, K.T. PreMetabo: An in silico phase I and II drug metabolism prediction platform. Drug Metab. Pharm. 2020, 35, 361–367. [Google Scholar] [CrossRef]Zaretzki, J.; Matlock, M.; Swamidass, S.J. XenoSite: Accurately predicting CYP-mediated sites of metabolism with neural networks. J. Chem. Inf. Model. 2013, 53, 3373–3383. [Google Scholar] [CrossRef] [PubMed]Rydberg, P.; Gloriam, D.E.; Zaretzki, J.; Breneman, C.; Olsen, L. SMARTCyp: A 2D Method for Prediction of Cytochrome P450-Mediated Drug Metabolism. ACS Med. Chem. Lett. 2010, 1, 96–100. [Google Scholar] [CrossRef] [PubMed]Olsen, L.; Montefiori, M.; Tran, K.P.; Jorgensen, F.S. SMARTCyp 3.0: Enhanced cytochrome P450 site-of-metabolism prediction server. Bioinformatics 2019, 35, 3174–3175. [Google Scholar] [CrossRef] [PubMed]Zhang, S.; Yan, Z.; Huang, Y.; Liu, L.; He, D.; Wang, W.; Fang, X.; Zhang, X.; Wang, F.; Wu, H.; et al. HelixADMET: A robust and endpoint extensible ADMET system incorporating self-supervised knowledge transfer. Bioinformatics 2022, 38, 3444–3453. [Google Scholar] [CrossRef]Wei, Y.; Li, S.; Li, Z.; Wan, Z.; Lin, J. Interpretable-ADMET: A Web Service for ADMET Prediction and Optimization based on Deep Neural Representation. Bioinformatics 2022, 38, 2863–2871. [Google Scholar] [CrossRef]Venkatraman, V. FP-ADMET: A compendium of fingerprint-based ADMET prediction models. J. Cheminform. 2021, 13. [Google Scholar] [CrossRef]Xiong, G.; Wu, Z.; Yi, J.; Fu, L.; Yang, Z.; Hsieh, C.; Yin, M.; Zeng, X.; Wu, C.; Lu, A.; et al. ADMETlab 2.0: An integrated online platform for accurate and comprehensive predictions of ADMET properties. Nucleic Acids Res. 2021, 49, W5–W14. [Google Scholar] [CrossRef]Yang, H.; Lou, C.; Sun, L.; Li, J.; Cai, Y.; Wang, Z.; Li, W.; Liu, G.; Tang, Y. admetSAR 2.0: Web-service for prediction and optimization of chemical ADMET properties. Bioinformatics 2019, 35, 1067–1069. [Google Scholar] [CrossRef] [PubMed]Daina, A.; Michielin, O.; Zoete, V. SwissADME: A free web tool to evaluate pharmacokinetics, drug-likeness and medicinal chemistry friendliness of small molecules. Sci. Rep. 2017, 7, 42717. [Google Scholar] [CrossRef] [PubMed]Wei, M.; Zhang, X.; Pan, X.; Wang, B.; Ji, C.; Qi, Y.; Zhang, J.Z. HobPre: Accurate prediction of human oral bioavailability for small molecules. J. Cheminform. 2022, 14, 1–10. [Google Scholar] [CrossRef] [PubMed]Li, X.; Xu, Y.J.; Lai, L.H.; Pei, J.F. Prediction of human cytochrome P450 inhibition using a multitask deep autoencoder neural network. Mol. Pharm. 2018, 15, 4336–4345. [Google Scholar] [CrossRef]Stork, C.; Embruch, G.; Šícho, M.; de Bruyn Kops, C.; Chen, Y.; Svozil, D.; Kirchmair, J. NERDD: A web portal providing access to in silico tools for drug discovery. Bioinformatics 2020, 36, 1291–1292. [Google Scholar] [CrossRef]Plonka, W.; Stork, C.; Sicho, M.; Kirchmair, J. CYPlebrity: Machine learning models for the prediction of inhibitors of cytochrome P450 enzymes. Bioorg. Med. Chem. 2021, 46, 116388. [Google Scholar] [CrossRef]Banerjee, P.; Dunkel, M.; Kemmler, E.; Preissner, R. SuperCYPsPred-a web server for the prediction of cytochrome activity. Nucleic Acids Res. 2020, 48, W580–W585. [Google Scholar] [CrossRef]Ridder, L.; Wagener, M. SyGMa: Combining expert knowledge and empirical scoring in the prediction of metabolites. ChemMedChem Chem. Enabling Drug Discov. 2008, 3, 821–832. [Google Scholar] [CrossRef]Nguyen-Vo, T.-H.; Trinh, Q.H.; Nguyen, L.; Nguyen-Hoang, P.-U.; Nguyen, T.-N.; Nguyen, D.T.; Nguyen, B.P.; Le, L. iCYP-MFE: Identifying human cytochrome P450 inhibitors using multitask learning and molecular fingerprint-embedded encoding. J. Chem. Inf. Model. 2021, 62, 5059–5068. [Google Scholar] [CrossRef]Shan, X.; Wang, X.; Li, C.-D.; Chu, Y.; Zhang, Y.; Xiong, Y.; Wei, D.-Q. Prediction of CYP450 enzyme–substrate selectivity based on the network-based label space division method. J. Chem. Inf. Model. 2019, 59, 4577–4586. [Google Scholar] [CrossRef]Park, H.; Brahma, R.; Shin, J.M.; Cho, K.H. Prediction of human cytochrome P450 inhibition using bio-selectivity induced deep neural network. Bulletin Korean Chem. Soc. 2022, 43, 261–269. [Google Scholar] [CrossRef]Dai, H.; Zheng, Y.-X.; Shan, X.-Q.; Chu, Y.-Y.; Wang, W.; Xiong, Y.; Wei, D.-Q. Computational prediction of the isoform specificity of cytochrome P450 substrates by an improved bayesian method. Res. Sq. 2019. [Google Scholar] [CrossRef]Raju, B.; Verma, H.; Narendra, G.; Sapra, B.; Silakari, O. Multiple machine learning, molecular docking, and ADMET screening approach for identification of selective inhibitors of CYP1B1. J. Biomol. Struct. Dyn. 2021, 1–16. [Google Scholar] [CrossRef] [PubMed]Shi, T.T.; Yang, Y.W.; Huang, S.H.; Chen, L.X.; Kuang, Z.Y.; Heng, Y.; Mei, H. Molecular image-based convolutional neural network for the prediction of ADMET properties. Chemom. Intell. Lab. Syst. 2019, 194, 103853. [Google Scholar] [CrossRef]Zhang, X.X.; Zhao, P.A.; Wang, Z.Y.; Xu, X.; Liu, G.X.; Tang, Y.; Li, W.H. In silico prediction of CYP2C8 inhibition with machine-learning methods. Chem. Res. Toxicol. 2021, 34, 1850–1859. [Google Scholar] [CrossRef]Racz, A.; Keseru, G.M. Large-scale evaluation of cytochrome P450 2C9 mediated drug interaction potential with machine learning-based consensus modeling. J. Comput.—Aided. Mol. Des. 2020, 34, 831–839. [Google Scholar] [CrossRef]Goldwaser, E.; Laurent, C.; Lagarde, N.; Fabrega, S.; Nay, L.; Villoutreix, B.O.; Jelsch, C.; Nicot, A.B.; Loriot, M.A.; Miteva, M.A. Machine learning-driven identification of drugs inhibiting cytochrome P450 2C9. PLoS Comput. Biol. 2022, 18, e1009820. [Google Scholar] [CrossRef]Zhao, J.; Liu, Y. Classification and prediction model of compound pharmacokinetic properties based on ensemble learning method. In Proceedings of the 2nd International Symposium on Artificial Intelligence for Medicine Sciences, Zhengzhou, China, 29–31 October 2021; pp. 526–531. [Google Scholar]Sasahara, K.; Shibata, M.; Sasabe, H.; Suzuki, T.; Takeuchi, K.; Umehara, K.; Kashiyama, E. Predicting drug metabolism and pharmacokinetics features of in-house compounds by a hybrid machine-learning model. Drug Metab. Pharmacokinet. 2021, 39. [Google Scholar] [CrossRef]Veith, H.; Southall, N.; Huang, R.; James, T.; Fayne, D.; Artemenko, N.; Shen, M.; Inglese, J.; Austin, C.P.; Lloyd, D.G. Comprehensive characterization of cytochrome P450 isozyme selectivity across chemical libraries. Nat. Biotechnol. 2009, 27, 1050–1055. [Google Scholar] [CrossRef]Shao, C.-Y.; Su, B.-H.; Tu, Y.-S.; Lin, C.; Lin, O.A.; Tseng, Y.J. Cyprules: A rule-based P450 inhibition prediction server. Bioinformatics 2015, 31, 1869–1871. [Google Scholar] [CrossRef]Su, B.-H.; Tu, Y.-S.; Lin, C.; Shao, C.-Y.; Lin, O.A.; Tseng, Y.J. Rule-based prediction models of cytochrome P450 inhibition. J. Chem. Inf. Model. 2015, 55, 1426–1434. [Google Scholar] [CrossRef] [PubMed]Preissner, S.; Kroll, K.; Dunkel, M.; Senger, C.; Goldsobel, G.; Kuzman, D.; Guenther, S.; Winnenburg, R.; Schroeder, M.; Preissner, R. SuperCYP: A comprehensive database on cytochrome P450 enzymes including a tool for analysis of CYP-drug interactions. Nucleic Acids Res. 2010, 38, D237–D243. [Google Scholar] [CrossRef] [PubMed]Fischer, M.; Knoll, M.; Sirim, D.; Wagner, F.; Funke, S.; Pleiss, J. The cytochrome P450 engineering database: A navigation and prediction tool for the cytochrome P450 protein family. Bioinformatics 2007, 23, 2015–2017. [Google Scholar] [CrossRef] [PubMed]Sirim, D.; Wagner, F.; Lisitsa, A.; Pleiss, J. The cytochrome P450 engineering database: Integration of biochemical properties. BMC Biochem. 2009, 10, 1–4. [Google Scholar] [CrossRef]Wu, Z.X.; Lei, T.L.; Shen, C.; Wang, Z.; Cao, D.S.; Hou, T.J. Admet evaluation in drug discovery. 19. Reliable prediction of human cytochrome P450 inhibition using artificial intelligence approaches. J. Chem. Inf. Model. 2019, 59, 4587–4601. [Google Scholar] [CrossRef]Backman, J.T.; Filppula, A.M.; Niemi, M.; Neuvonen, P.J. Role of cytochrome P450 2C8 in drug metabolism and interactions. Pharmacol. Rev. 2016, 68, 168–241. [Google Scholar] [CrossRef]Walsky, R.L.; Gaman, E.A.; Obach, R.S. Examination of 209 drugs for inhibition of cytochrome P450 2C8. J. Clin. Pharmacol. 2005, 45, 68–78. [Google Scholar] [CrossRef]Wenzel, J.; Matter, H.; Schmidt, F. Predictive multitask deep neural network models for adme-tox properties: Learning from large data sets. J. Chem. Inf. Model. 2019, 59, 1253–1268. [Google Scholar] [CrossRef]Liu, K.; Sun, X.; Jia, L.; Ma, J.; Xing, H.; Wu, J.; Gao, H.; Sun, Y.; Boulnois, F.; Fan, J. Chemi-net: A molecular graph convolutional network for accurate drug property prediction. Int. J. Mol. Sci. 2019, 20, 3389. [Google Scholar] [CrossRef]Dong, J.; Wang, Z.; Cui, J.; Meng, Q.; Li, S. Synthesis and structure-activity relationship studies of α-naphthoflavone derivatives as CYP1b1 inhibitors. Eur. J. Med. Chem. 2020, 187, 111938. [Google Scholar] [CrossRef]Kubo, M.; Yamamoto, K.; Itoh, T. Design and synthesis of selective CYP1B1 inhibitor via dearomatization of α-naphthoflavone. Bioorganic Med. Chemistry. 2019, 27, 285–304. [Google Scholar] [CrossRef] [PubMed]Meng, Q.; Wang, Z.; Cui, J.; Cui, Q.; Dong, J.; Zhang, Q.; Li, S. Design, synthesis, and biological evaluation of cytochrome P450 1b1 targeted molecular imaging probes for colorectal tumor detection. J. Med. Chem. 2018, 61, 10901–10909. [Google Scholar] [CrossRef] [PubMed]Czechtizky, W.; Su, W.; Ripa, L.; Schiesser, S.; Höijer, A.; Cox, R.J. Advances in the design of new types of inhaled medicines. Prog. Med. Chem. 2022, 61, 93–162. [Google Scholar] [PubMed]Smith, P.; Sorich, M.; Low, L.; McKinnon, R.; Miners, J. Towards integrated ADME prediction: Past, present and future directions for modelling metabolism by UDP-glucuronosyltransferases. J. Mol. Graph. Model. 2004, 22, 507–517. [Google Scholar] [CrossRef]Mazzolari, A.; Afzal, A.M.; Pedretti, A.; Testa, B.; Vistoli, G.; Bender, A. Prediction of UGT-mediated metabolism using the manually curated metaqsar database. ACS Med. Chem. Lett. 2019, 10, 633–638. [Google Scholar] [CrossRef]Pedretti, A.; Mazzolari, A.; Vistoli, G.; Testa, B. Metaqsar: An integrated database engine to manage and analyze metabolic data. J. Med. Chem. 2018, 61, 1019–1030. [Google Scholar] [CrossRef]Cai, Y.; Yang, H.; Li, W.; Liu, G.; Lee, P.W.; Tang, Y. Computational prediction of site of metabolism for UGT-catalyzed reactions. J. Chem. Inf. Model. 2019, 59, 1085–1095. [Google Scholar] [CrossRef]Lee, P.W.; Aizawa, H.; Gan, L.; Prakash, C.; Zhong, D. Handbook of Metabolic Pathways of Xenobiotics; Wiley Online Library: Hoboken, NJ, USA, 2014. [Google Scholar]Peng, J.; Lu, J.; Shen, Q.; Zheng, M.; Luo, X.; Zhu, W.; Jiang, H.; Chen, K. In silico site of metabolism prediction for human UGT-catalyzed reactions. Bioinformatics 2014, 30, 398–405. [Google Scholar] [CrossRef]Rudik, A.; Dmitriev, A.; Lagunin, A.; Filimonov, D.; Poroikov, V. SOMP: Web server for in silico prediction of sites of metabolism for drug-like compounds. Bioinformatics 2015, 31, 2046–2048. [Google Scholar] [CrossRef]Kanehisa, M.; Furumichi, M.; Tanabe, M.; Sato, Y.; Morishima, K. KEGG: New perspectives on genomes, pathways, diseases and drugs. Nucleic Acids Res. 2017, 45, D353–D361. [Google Scholar] [CrossRef]Horde, G.W.; Gupta, V. Drug clearance. In Statpearls; Treasure Island: FL, USA, 2022. [Google Scholar]Wade, K.C. Pharmacokinetics in neonatal medicine. In Fanaroff and Martin’s Neonatal-Perinatal Medicine; Elsevier: Philadelphia, PA, USA, 2020; pp. 722–734. [Google Scholar]Smith, D.A.; Beaumont, K.; Maurer, T.S.; Di, L. Clearance in drug design. J. Med. Chem. 2019, 62, 2245–2255. [Google Scholar] [CrossRef] [PubMed]Esaki, T.; Watanabe, R.; Kawashima, H.; Ohashi, R.; Natsume-Kitatani, Y.; Nagao, C.; Mizuguchi, K. Data curation can improve the prediction accuracy of metabolic intrinsic clearance. Mol. Inform. 2019, 38, 1800086. [Google Scholar] [CrossRef] [PubMed]Hsiao, Y.W.; Fagerholm, U.; Norinder, U. In silico categorization of in vivo intrinsic clearance using machine learning. Mol. Pharm. 2013, 10, 1318–1321. [Google Scholar] [CrossRef]Chen, J.H.; Yang, H.B.; Zhu, L.; Wu, Z.R.; Li, W.H.; Tang, Y.; Liu, G.X. In silico prediction of human renal clearance of compounds using quantitative structure-pharmacokinetic relationship models. Chem. Res. Toxicol. 2020, 33, 640–650. [Google Scholar] [CrossRef] [PubMed]Wang, Y.C.; Liu, H.C.; Fan, Y.R.; Chen, X.Y.; Yang, Y.; Zhu, L.; Zhao, J.N.; Chen, Y.D.; Zhang, Y.M. In silico prediction of human intravenous pharmacokinetic parameters with improved accuracy. J. Chem. Inf. Model. 2019, 59, 3968–3980. [Google Scholar] [CrossRef]Kosugi, Y.; Hosea, N. Direct comparison of total clearance prediction: Computational machine learning model versus bottom-up approach using in vitro assay. Mol. Pharm. 2020, 17, 2299–2309. [Google Scholar] [CrossRef]Watanabe, R.; Ohashi, R.; Esaki, T.; Kawashima, H.; Natsume-Kitatani, Y.; Nagao, C.; Mizuguchi, K. Development of an in silico prediction system of human renal excretion and clearance from chemical structure information incorporating fraction unbound in plasma as a descriptor. Sci. Rep. 2019, 9, 18782. [Google Scholar] [CrossRef] [PubMed]Oprisiu, I.; Winiwarter, S. Silico ADME Modeling; Academic Press: Cambridge, MA, USA, 2021; pp. 208–222. [Google Scholar]Mamada, H.; Nomura, Y.; Uesawa, Y. Prediction model of clearance by a novel quantitative structure-activity relationship approach, combination deepsnap-deep learning and conventional machine learning. ACS Omega 2021, 6, 23570–23577. [Google Scholar] [CrossRef]Sohlenius-Sternbeck, A.-K.; Terelius, Y. Evaluation of ADMET predictor in early discovery drug metabolism and pharmacokinetics project work. Drug Metab. Dispos. 2022, 50, 95–104. [Google Scholar] [CrossRef]Lombardo, F.; Berellini, G.; Obach, R.S. Trend analysis of a database of intravenous pharmacokinetic parameters in humans for 1352 drug compounds. Drug Metab. Dispos. 2018, 46, 1466–1477. [Google Scholar] [CrossRef]Paine, S.W.; Barton, P.; Bird, J.; Denton, R.; Menochet, K.; Smith, A.; Tomkinson, N.P.; Chohan, K.K. A rapid computational filter for predicting the rate of human renal clearance. J. Mol. Graph. Model. 2010, 29, 529–537. [Google Scholar] [CrossRef] [PubMed]Varma, M.V.; Feng, B.; Obach, R.S.; Troutman, M.D.; Chupka, J.; Miller, H.R.; El-Kattan, A. Physicochemical determinants of human renal clearance. J. Med. Chem. 2009, 52, 4844–4852. [Google Scholar] [CrossRef]Lombardo, F.; Obach, R.S.; Varma, M.V.; Stringer, R.; Berellini, G. Clearance mechanism assignment and total clearance prediction in human based upon in silico models. J. Med. Chem. 2014, 57, 4397–4405. [Google Scholar] [CrossRef] [PubMed]Scotcher, D.; Jones, C.; Rostami-Hodjegan, A.; Galetin, A. Novel minimal physiologically-based model for the prediction of passive tubular reabsorption and renal excretion clearance. Eur. J. Pharm. Sci. 2016, 94, 59–71. [Google Scholar] [CrossRef]Kanehisa, M.; Goto, S.; Hattori, M.; Aoki-Kinoshita, K.F.; Itoh, M.; Kawashima, S.; Katayama, T.; Araki, M.; Hirakawa, M. From genomics to chemical genomics: New developments in KEGG. Nucleic Acids Res. 2006, 34, D354–D357. [Google Scholar] [CrossRef] [PubMed]Varma, M.V.; Obach, R.S.; Rotter, C.; Miller, H.R.; Chang, G.; Steyn, S.J.; El-Kattan, A.; Troutman, M.D. Physicochemical space for optimum oral bioavailability: Contribution of human intestinal absorption and first-pass elimination. J. Med. Chem. 2010, 53, 1098–1108. [Google Scholar] [CrossRef] [PubMed]Mora, A.M.; Subramanian, V.; Miljković, F. Multi-task convolutional neural networks for predicting in vitro clearance endpoints from molecular images. J. Comput.—Aided Mol. Des. 2022, 36, 443–457. [Google Scholar] [CrossRef]Hallare, J.; Gerriets, V. Half Life; Statpearls: Treasure Island, FL, USA, 2022. [Google Scholar]Smith, D.A.; Beaumont, K.; Maurer, T.S.; Di, L. Relevance of half-life in drug design. J. Med. Chem. 2018, 61, 4273–4282. [Google Scholar] [CrossRef]Podlewska, S.; Kafel, R. Metstabon-online platform for metabolic stability predictions. Int. J. Mol. Sci. 2018, 19, 1040. [Google Scholar] [CrossRef]Wishart, D.S.; Guo, A.; Oler, E.; Wang, F.; Anjum, A.; Peters, H.; Dizon, R.; Sayeeda, Z.; Tian, S.; Lee, B.L.; et al. HMDB 5.0: The human metabolome database for 2022. Nucleic Acids Res. 2022, 50, D622–D631. [Google Scholar] [CrossRef]Montenegro-Burke, J.R.; Guijas, C.; Siuzdak, G. Metlin: A tandem mass spectral library of standards. Comput. Methods Data Anal. Metab. 2020, 2104, 149–163. [Google Scholar] [CrossRef]Haug, K.; Cochrane, K.; Nainala, V.C.; Williams, M.; Chang, J.; Jayaseelan, K.V.; O’Donovan, C. MetaboLights: A resource evolving in response to the needs of its scientific community. Nucleic Acids Res. 2020, 48, D440–D444. [Google Scholar] [CrossRef] [PubMed]Kanehisa, M. ‘In Silico‘ Simulation of Biological Processes: Novartis Foundation Symposium 247. In The Kegg Database; Wiley Online Library: New York, NY, USA, 2002; pp. 91–103. [Google Scholar]Trupp, M.; Altman, T.; Fulcher, C.A.; Caspi, R.; Krummenacker, M.; Paley, S.; Karp, P.D. Beyond the genome (btg) is a (pgdb) pathway genome database: HumanCyc. Genome Biol. 2010, 11, 1. [Google Scholar] [CrossRef]Norsigian, C.J.; Pusarla, N.; McConn, J.L.; Yurkovich, J.T.; Dräger, A.; Palsson, B.O.; King, Z. Bigg models 2020: Multi-strain genome-scale models and expansion across the phylogenetic tree. Nucleic Acids Res. 2020, 48, D402–D406. [Google Scholar] [CrossRef]Wishart, D.S.; Feunang, Y.D.; Guo, A.C.; Lo, E.J.; Marcu, A.; Grant, J.R.; Sajed, T.; Johnson, D.; Li, C.; Sayeeda, Z. DrugBank 5.0: A major update to the drugbank database for 2018. Nucleic Acids Res. 2018, 46, D1074–D1082. [Google Scholar] [CrossRef]Mendez, D.; Gaulton, A.; Bento, A.P.; Chambers, J.; De Veij, M.; Félix, E.; Magariños, M.P.; Mosquera, J.F.; Mutowo, P.; Nowotka, M. ChEMBL: Towards direct deposition of bioassay data. Nucleic Acids Res. 2019, 47, D930–D940. [Google Scholar] [CrossRef]Pence, H.E.; Williams, A. ChemSpider: An online chemical information resource. J. Chem. Educ. 2010, 87, 1123–1124. [Google Scholar] [CrossRef]Kim, S.; Chen, J.; Cheng, T.; Gindulyte, A.; He, J.; He, S.; Li, Q.; Shoemaker, B.A.; Thiessen, P.A.; Yu, B. PubChem 2019 update: Improved access to chemical data. Nucleic Acids Res. 2019, 47, D1102–D1109. [Google Scholar] [CrossRef]Irwin, J.J.; Tang, K.G.; Young, J.; Dandarchuluun, C.; Wong, B.R.; Khurelbaatar, M.; Moroz, Y.S.; Mayfield, J.; Sayle, R.A. ZINC20—A free ultralarge-scale chemical database for ligand discovery. J. Chem. Inf. Model. 2020, 60, 6065–6073. [Google Scholar] [CrossRef]Sushko, I.; Novotarskyi, S.; Körner, R.; Pandey, A.K.; Rupp, M.; Teetz, W.; Brandmaier, S.; Abdelaziz, A.; Prokopenko, V.V.; Tanchuk, V.Y. Online chemical modeling environment (OCHEM): Web platform for data storage, model development and publishing of chemical information. J. Comput.—Aided Mol. Des. 2011, 25, 533–554. [Google Scholar] [CrossRef]Huang, K.; Fu, T.; Gao, W.; Zhao, Y.; Roohani, Y.; Leskovec, J.; Coley, C.W.; Xiao, C.; Sun, J.; Zitnik, M. Artificial intelligence foundation for therapeutic science. Nat. Chem. Biol. 2022, 18, 1033–1036. [Google Scholar] [CrossRef] [PubMed]Kass-Hout, T.A.; Xu, Z.; Mohebbi, M.; Nelsen, H.; Baker, A.; Levine, J.; Johanson, E.; Bright, R.A. OpenFDA: An innovative platform providing access to a wealth of fda’s publicly available data. J. Am. Med. Inform. Assoc. 2016, 23, 596–600. [Google Scholar] [CrossRef] [PubMed]Bermingham, K.M.; Brennan, L.; Segurado, R.; Barron, R.E.; Gibney, E.R.; Ryan, M.F.; Gibney, M.J.; O’Sullivan, A.M. Genetic and environmental contributions to variation in the stable urinary NMR metabolome over time: A classic twin study. J. Proteome Res. 2021, 20, 3992–4000. [Google Scholar] [CrossRef]Pearson, P.G.; Wienkers, L.C. Handbook of Drug Metabolism; CRC Press: Boca Raton, FL, USA, 2019. [Google Scholar]Arabi, A.A. Artificial intelligence in drug design: Algorithms, applications, challenges and ethics. Future Drug Discov. 2021, 3, FDD59. [Google Scholar] [CrossRef]Jiménez-Luna, J.; Grisoni, F.; Schneider, G. Drug discovery with explainable artificial intelligence. Nat. Mach. Intell. 2020, 2, 573–584. [Google Scholar] [CrossRef]Aittokallio, T. What are the current challenges for machine learning in drug discovery and repurposing? Expert Opin. Drug Discov. 2022, 17, 423–425. [Google Scholar] [CrossRef]Sheridan, R.P. Interpretation of QSAR models by coloring atoms according to changes in predicted activity: How robust is it? J. Chem. Inf. Model. 2019, 59, 1324–1337. [Google Scholar] [CrossRef] [PubMed]Azram, N.A.; Atan, R.; Mustafa, S.; Desa, M.N.M. A review on integration of scientific experimental data through metadata. Recent Trends Adv. Wirel. IoT-Enabled Netw. 2019, 63–72. [Google Scholar] [CrossRef]Subramanian, I.; Verma, S.; Kumar, S.; Jere, A.; Anamika, K. Multi-omics data integration, interpretation, and its application. Bioinform. Biol. Insights 2020, 14, 1177932219899051. [Google Scholar] [CrossRef]
Figure 1. Overview of drug metabolism and excretion. Figure 1. Overview of drug metabolism and excretion. 
Figure 2. Phase I and phase II in drug metabolism. Figure 2. Phase I and phase II in drug metabolism. Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.© 2023 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/). Share and CiteMDPI and ACS StyleTran, T.T.V.; Tayara, H.; Chong, K.T.Artificial Intelligence in Drug Metabolism and Excretion Prediction: Recent Advances, Challenges, and Future Perspectives. Pharmaceutics 2023, 15, 1260.https://doi.org/10.3390/pharmaceutics15041260
AMA StyleTran TTV, Tayara H, Chong KT.Artificial Intelligence in Drug Metabolism and Excretion Prediction: Recent Advances, Challenges, and Future Perspectives. Pharmaceutics. 2023; 15(4):1260.https://doi.org/10.3390/pharmaceutics15041260
Chicago/Turabian StyleTran, Thi Tuyet Van, Hilal Tayara, and Kil To Chong.2023. "Artificial Intelligence in Drug Metabolism and Excretion Prediction: Recent Advances, Challenges, and Future Perspectives" Pharmaceutics 15, no. 4: 1260.https://doi.org/10.3390/pharmaceutics15041260
Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.Article MetricsNoNoArticle Access StatisticsFor more information on the journal statistics, click here.Multiple requests from the same IP address are counted as one view.Zoom| Orient | As Lines | As Sticks | As Cartoon | As Surface |Previous Scene|Next SceneCiteExport citation file: BibTeX |EndNote |RISMDPI and ACS StyleTran, T.T.V.; Tayara, H.; Chong, K.T.Artificial Intelligence in Drug Metabolism and Excretion Prediction: Recent Advances, Challenges, and Future Perspectives. Pharmaceutics 2023, 15, 1260.https://doi.org/10.3390/pharmaceutics15041260
AMA StyleTran TTV, Tayara H, Chong KT.Artificial Intelligence in Drug Metabolism and Excretion Prediction: Recent Advances, Challenges, and Future Perspectives. Pharmaceutics. 2023; 15(4):1260.https://doi.org/10.3390/pharmaceutics15041260
Chicago/Turabian StyleTran, Thi Tuyet Van, Hilal Tayara, and Kil To Chong.2023. "Artificial Intelligence in Drug Metabolism and Excretion Prediction: Recent Advances, Challenges, and Future Perspectives" Pharmaceutics 15, no. 4: 1260.https://doi.org/10.3390/pharmaceutics15041260
Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.clear Pharmaceutics,EISSN 1999-4923,Published by MDPIRSSContent AlertFurther InformationArticle Processing ChargesPay an InvoiceOpen Access PolicyContact MDPIJobs at MDPIGuidelinesFor AuthorsFor ReviewersFor EditorsFor LibrariansFor PublishersFor SocietiesFor Conference OrganizersMDPI InitiativesSciforumMDPI BooksPreprints.orgScilitSciProfilesEncyclopediaJAMSProceedings SeriesFollow MDPILinkedInFacebookTwitter
© 1996-2024 MDPI (Basel, Switzerland) unless otherwise statedDisclaimerDisclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.Terms and ConditionsPrivacy PolicyWe use cookies on our website to ensure you get the best experience.Read more about our cookies here.AcceptShare LinkCopyclearSharehttps://www.mdpi.com/2251230clearBack to TopTop